diff --git a/Leecode/src/main/java/com/markilue/leecode/backtrace/PermuteUnique.java b/Leecode/src/main/java/com/markilue/leecode/backtrace/PermuteUnique.java new file mode 100644 index 0000000..ae7369a --- /dev/null +++ b/Leecode/src/main/java/com/markilue/leecode/backtrace/PermuteUnique.java @@ -0,0 +1,120 @@ +package com.markilue.leecode.backtrace; + +import org.junit.Test; + +import java.util.ArrayList; +import java.util.Arrays; +import java.util.HashSet; +import java.util.List; + +/** + * @BelongsProject: Leecode + * @BelongsPackage: com.markilue.leecode.backtrace + * @Author: markilue + * @CreateTime: 2022-10-20 09:46 + * @Description: TODO 力扣47题 全排列II： + * 给定一个可包含重复数字的序列 nums ，按任意顺序 返回所有不重复的全排列。 + * @Version: 1.0 + */ +public class PermuteUnique { + + @Test + public void test() { + int[] nums = {1, 1, 2}; + System.out.println(permuteUnique(nums)); + } + + @Test + public void test1() { + int[] nums = {1, 2, 3}; + System.out.println(permuteUnique(nums)); + } + + List> result = new ArrayList<>(); + List cur = new ArrayList<>(); + int[] index = new int[10]; + + /** + * 思路：要求不重复且不要求顺序，则还是两种方式： + * 1）第一种即排序之后使用used数组记录是否用过，同一树层不能使用相同的数据，不同树层可以使用相同的数; + * 2)使用map记录数字次数 + * 这里尝试使用1），但使用set了 + * 速度击败44.42%，内存击败39.13% + * + * @param nums + * @return + */ + public List> permuteUnique(int[] nums) { + + boolean[] used = new boolean[nums.length]; + Arrays.sort(nums); + backtracking(nums, used); + +// backtracking(nums); + return result; + + + } + + public void backtracking(int[] nums) { + +// if(start>nums.length){ +// return; +// } + + if (cur.size() == nums.length) { + result.add(new ArrayList<>(cur)); + return; + } + + HashSet set = new HashSet<>(); + for (int i = 0; i < nums.length; i++) { + + if (set.contains(nums[i]) || index[i] == 1) { + continue; + } + index[i] = 1; + set.add(nums[i]); + cur.add(nums[i]); + backtracking(nums); + cur.remove(cur.size() - 1); + index[i] = 0; + + } + + } + + + /** + * 速度击败99.85%，内存击败89.51% + * @param nums + * @param used + */ + public void backtracking(int[] nums, boolean[] used) { + + + if (cur.size() == nums.length) { + result.add(new ArrayList<>(cur)); + return; + } + + for (int i = 0; i < nums.length; i++) { + + if ((i > 0 && nums[i] == nums[i - 1] && used[i - 1] == false)) { + continue; + } + + if (used[i] == false) { +// index[i] = 1; + used[i] = true; + cur.add(nums[i]); + backtracking(nums,used); + cur.remove(cur.size() - 1); +// index[i] = 0; + used[i] = false; + } + + } + + } +} diff --git a/Leecode/src/main/java/com/markilue/leecode/backtrace/SolveNQueens.java b/Leecode/src/main/java/com/markilue/leecode/backtrace/SolveNQueens.java new file mode 100644 index 0000000..1ca63be --- /dev/null +++ b/Leecode/src/main/java/com/markilue/leecode/backtrace/SolveNQueens.java @@ -0,0 +1,237 @@ +package com.markilue.leecode.backtrace; + +import com.sun.crypto.provider.PBEWithMD5AndDESCipher; +import org.junit.Test; + +import java.util.*; + +/** + * @BelongsProject: Leecode + * @BelongsPackage: com.markilue.leecode.backtrace + * @Author: markilue + * @CreateTime: 2022-10-20 11:14 + * @Description: TODO 力扣51题 N皇后： + * 按照国际象棋的规则，皇后可以攻击与之处在同一行或同一列或同一斜线上的棋子。 + * n 皇后问题 研究的是如何将 n 个皇后放置在 n×n 的棋盘上，并且使皇后彼此之间不能相互攻击。 + * 给你一个整数 n ，返回所有不同的 n 皇后问题 的解决方案。 + * 每一种解法包含一个不同的 n 皇后问题 的棋子放置方案，该方案中 'Q' 和 '.' 分别代表了皇后和空位。 + * @Version: 1.0 + */ +public class SolveNQueens { + + @Test + public void test() { +// StringBuilder builder=new StringBuilder("1"); +// System.out.println(builder.delete(0, builder.length()-1)); + + System.out.println(solveNQueens(5)); + } + @Test + public void test1() { +// StringBuilder builder=new StringBuilder("1"); +// System.out.println(builder.delete(0, builder.length()-1)); + + System.out.println(solveNQueens(8).size()); + } + + + /** + * 主要是注意条件，同行同列斜线都不行 + * 速度击败39.74%，内存击败92.94% + * @param n + * @return + */ + public List> solveNQueens(int n) { + int[] used = new int[n]; +// for (int i = 0; i < n; i++) { +// used[i]=Integer.MIN_VALUE; +// } + backtracking1(n, used, 0); + + + return result; + } + + + List> result = new ArrayList<>(); + List cur = new ArrayList<>(); + + + /** + * used数组记录位置使用情况，used[i]=k表示第i行第k列被占用 + * value表示当前是在放置第多少列了 + * 效率低下原因，for循环需要检查前面所有的列，效率太低下了：是否可以拿一个set记录一下不能放的位置？ + * @param n + * @param used + */ + public void backtracking(int n, int[] used, int value) { + + if (cur.size() == n) { + result.add(new ArrayList<>(cur)); + return; + } + StringBuilder stringBuilder = new StringBuilder(); + + //横向行走 + for (int i = 0; i < n; i++) { + + boolean flag = true; + //不合规则 + for (int j = 0; j < value; j++) { + if(!check(j,used[j],value,i)){ + flag=false; + break; + } + } + + if (!flag) { + continue; + } + used[value] = i; + + for (int j = 0; j < i; j++) { + stringBuilder.append("."); + } + stringBuilder.append("Q"); + for (int j = i + 1; j < n; j++) { + stringBuilder.append("."); + } + cur.add(stringBuilder.toString()); + //纵向行走 + backtracking(n, used, value + 1); + cur.remove(cur.size() - 1); + stringBuilder.delete(0, stringBuilder.length()); + } + + } + + /** + * used数组记录位置使用情况，used[i]=k表示第i行第k列被占用 + * value表示当前是在放置第多少列了 + * 优化stringbuilder,使用replace方法 + * 好像效率反而降低了 + * @param n + * @param used + */ + public void backtracking1(int n, int[] used, int value) { + + if (cur.size() == n) { + result.add(new ArrayList<>(cur)); + return; + } + + StringBuilder build=new StringBuilder(); + for (int i = 0; i < n; i++) { + build.append("."); + } + + + StringBuilder stringBuilder = new StringBuilder(build); + + //横向行走 + for (int i = 0; i < n; i++) { + + boolean flag = true; + //不合规则 + for (int j = 0; j < value; j++) { + if(!check(j,used[j],value,i)){ + flag=false; + break; + } + } + + if (!flag) { + continue; + } + used[value] = i; + stringBuilder.replace(i,i+1,"Q"); + cur.add(stringBuilder.toString()); + //纵向行走 + backtracking(n, used, value + 1); + cur.remove(cur.size() - 1); + stringBuilder.replace(i,i+1,"."); +// stringBuilder.delete(i, stringBuilder.length()); + } + + } + + + public boolean check(int i,int value1,int j,int value2){ + + //同列 + if(value1==value2){ + return false; + } + //斜线 + if(Math.abs(i-j)== Math.abs(value1-value2)){ + return false; + } + + return true; + + + } + + + /** + * 官方代码：利用set记录不能使用的行列。 + * 效率好像也不高：击败39.74% + * @param n + * @return + */ + public List> solveNQueens1(int n) { + List> solutions = new ArrayList>(); + int[] queens = new int[n]; + Arrays.fill(queens, -1); + Set columns = new HashSet(); + Set diagonals1 = new HashSet(); + Set diagonals2 = new HashSet(); + backtrack(solutions, queens, n, 0, columns, diagonals1, diagonals2); + return solutions; + } + + public void backtrack(List> solutions, int[] queens, int n, int row, Set columns, Set diagonals1, Set diagonals2) { + if (row == n) { + List board = generateBoard(queens, n); + solutions.add(board); + } else { + for (int i = 0; i < n; i++) { + //竖着的 + if (columns.contains(i)) { + continue; + } + //斜线 + int diagonal1 = row - i; + if (diagonals1.contains(diagonal1)) { + continue; + } + int diagonal2 = row + i; + if (diagonals2.contains(diagonal2)) { + continue; + } + queens[row] = i; + columns.add(i); + diagonals1.add(diagonal1); + diagonals2.add(diagonal2); + backtrack(solutions, queens, n, row + 1, columns, diagonals1, diagonals2); + queens[row] = -1; + columns.remove(i); + diagonals1.remove(diagonal1); + diagonals2.remove(diagonal2); + } + } + } + + public List generateBoard(int[] queens, int n) { + List board = new ArrayList(); + for (int i = 0; i < n; i++) { + char[] row = new char[n]; + Arrays.fill(row, '.'); + row[queens[i]] = 'Q'; + board.add(new String(row)); + } + return board; + } + + +} diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-3.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-3.py new file mode 100644 index 0000000..6525dc7 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-3.py @@ -0,0 +1,711 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_3 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-34.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-34.py new file mode 100644 index 0000000..e4f06f2 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-34.py @@ -0,0 +1,714 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_34 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-35.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-35.py new file mode 100644 index 0000000..ca132c2 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-35.py @@ -0,0 +1,714 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_35 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-4.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-4.py new file mode 100644 index 0000000..924cb1b --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-4.py @@ -0,0 +1,714 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_4 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-45.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-45.py new file mode 100644 index 0000000..dfd1cb3 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-45.py @@ -0,0 +1,714 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_45 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-5.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-5.py new file mode 100644 index 0000000..2ed6c52 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-5.py @@ -0,0 +1,714 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:55 +@Usage : +@Desc : RNet直接进行分类 +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_5 import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_C" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + # output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False,isSave:bool=True): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + + #误报率，漏报率，准确性的计算 + + + if isSave: + + np.savetxt(save_mse_name, total_result, delimiter=',') + if isPlot: + plt.figure(1, figsize=(6.0, 2.68)) + plt.subplots_adjust(left=0.1, right=0.94, bottom=0.2, top=0.9, wspace=None, + hspace=None) + plt.tight_layout() + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + plt.text(test_data.shape[0] * 2 / 3+1000, 0.7, "Truth Fault", fontsize=10, color='red', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xticks(range(6),('06/09/17','12/09/17','18/09/17','24/09/17','29/09/17')) # 设置x轴的标尺 + plt.tick_params() #设置轴显示 + plt.xlabel("time",fontdict=font1) + plt.ylabel("confience",fontdict=font1) + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10},fontdict=font1) + + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + # train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + # step_one_model = Joint_Monitoring() + # # step_one_model.load_weights(save_name) + # # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + #### TODO 第二步训练 + ### healthy_data.shape: (300333,120,10) + ### unhealthy_data.shape: (16594,10) + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, + # train_data=train_data, + # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + ### TODO 测试测试集 + # step_one_model = Joint_Monitoring() + # step_one_model.load_weights(save_name) + step_two_model = Joint_Monitoring() + step_two_model.load_weights(save_step_two_name) + # test(step_one_model=step_one_model, step_two_model=step_two_model, test_data=test_data, test_label1=test_label1, + # test_label2=np.expand_dims(test_label2, axis=-1)) + + ###TODO 展示全部的结果 + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + showResult(step_two_model, test_data=all_data, isPlot=True) + + pass \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-C.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-C.py index db53ae3..f855bd7 100644 --- a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-C.py +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-C.py @@ -675,11 +675,11 @@ if __name__ == '__main__': ### unhealthy_data.shape: (16594,10) # healthy_size, _, _ = train_data_healthy.shape # unhealthy_size, _, _ = train_data_unhealthy.shape - # train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( - # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], - # healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], - # healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, - # unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) # train_step_two(step_one_model=step_one_model, step_two_model=step_two_model, # train_data=train_data, # train_label1=train_label1, train_label2=np.expand_dims(train_label2, axis=-1)) diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-S.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-S.py new file mode 100644 index 0000000..4b046c0 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-S.py @@ -0,0 +1,687 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:53 +@Usage : +@Desc : Rnet-SE模型对比 +''' + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention +from condition_monitoring.data_deal import loadData +from model.Joint_Monitoring.compare.RNet_S import Joint_Monitoring + +from model.CommonFunction.CommonFunction import * +from sklearn.model_selection import train_test_split +from tensorflow.keras.models import load_model, save_model +import random + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "RNet_S" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' +# save_name = "E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\hard_model\weight\joint_timestamp120_feature10_weight_epoch11_0.0077/weight" +save_name = "./model/weight/{0}/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "../hard_model/two_weight/{0}_timestamp{1}_feature{2}_weight_epoch14/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_S/{0}_timestamp{1}_feature{2}_mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_S/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +# save_name = "../model/joint/{0}_timestamp{1}_feature{2}.h5".format(model_name, +# time_stamp, +# feature_num, +# batch_size, +# EPOCH) +# save_step_two_name = "../model/joint_two/{0}_timestamp{1}_feature{2}.h5".format(model_name, +# time_stamp, +# feature_num, +# batch_size, +# EPOCH) +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + + +def remove(data, time_stamp=time_stamp): + rows, cols = data.shape + print("remove_data.shape:", data.shape) + num = int(rows / time_stamp) + + return data[:num * time_stamp, :] + pass + + +# 不重叠采样 +def get_training_data(data, time_stamp: int = time_stamp): + removed_data = remove(data=data) + rows, cols = removed_data.shape + print("removed_data.shape:", data.shape) + print("removed_data:", removed_data) + train_data = np.reshape(removed_data, [-1, time_stamp, cols]) + print("train_data:", train_data) + batchs, time_stamp, cols = train_data.shape + + for i in range(1, batchs): + each_label = np.expand_dims(train_data[i, 0, :], axis=0) + if i == 1: + train_label = each_label + else: + train_label = np.concatenate([train_label, each_label], axis=0) + + print("train_data.shape:", train_data.shape) + print("train_label.shape", train_label.shape) + return train_data[:-1, :], train_label + + +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +# RepConv重参数化卷积 +def RepConv(input_tensor, k=3): + _, _, output_dim = input_tensor.shape + conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=k, strides=1, padding='SAME')(input_tensor) + b1 = tf.keras.layers.BatchNormalization()(conv1) + + conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='SAME')(input_tensor) + b2 = tf.keras.layers.BatchNormalization()(conv2) + + b3 = tf.keras.layers.BatchNormalization()(input_tensor) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + return out + + +# RepBlock模块 +def RepBlock(input_tensor, num: int = 3): + for i in range(num): + input_tensor = RepConv(input_tensor) + return input_tensor + + +# GAP 全局平均池化 +def Global_avg_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + output = tf.multiply(input_tensor, s1) + return output + + +# GDP 全局动态池化 +def Global_Dynamic_channelAttention(input_tensor): + _, length, channel = input_tensor.shape + DWC1 = DepthwiseConv1D(kernel_size=1, padding='SAME')(input_tensor) + + # GAP + GAP = tf.keras.layers.GlobalAvgPool1D()(DWC1) + c1 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GAP) + s1 = tf.nn.sigmoid(c1) + + # GMP + GMP = tf.keras.layers.GlobalMaxPool1D()(DWC1) + c2 = tf.keras.layers.Conv1D(filters=channel, kernel_size=1, padding='SAME')(GMP) + s3 = tf.nn.sigmoid(c2) + + output = tf.multiply(input_tensor, s1) + return output + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +# 正则化 +def Regularization(data): + rows, cols = data.shape + print("正则化之前:", data) + print(data.shape) + print("======================") + + # 正则化 + mean = np.mean(data, axis=0) + mean = np.broadcast_to(mean, shape=[rows, cols]) + dst = np.sqrt(np.var(data, axis=0)) + dst = np.broadcast_to(dst, shape=[rows, cols]) + data = (data - mean) / dst + print("正则化之后:", data) + print(data.shape) + + return data + pass + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + t = 0 + mid = np.mean(data, axis=0) + standard = np.sqrt(np.var(data, axis=0)) + UCL = mid + K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + LCL = mid - K * standard * np.sqrt(namuda / (2 - namuda) * (1 - (1 - namuda) ** 2 * t)) + return mid, UCL, LCL + pass + + + + + +def condition_monitoring_model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + conv1 = tf.keras.layers.Conv1D(filters=256, kernel_size=1)(input) + GRU1 = tf.keras.layers.GRU(128, return_sequences=False)(conv1) + d1 = tf.keras.layers.Dense(300)(GRU1) + output = tf.keras.layers.Dense(10)(d1) + + model = tf.keras.Model(inputs=input, outputs=output) + + return model + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# trian_data:(300455,120,10) +# trian_label1:(300455,10) +# trian_label2:(300455,) +def train_step_one(train_data, train_label1, train_label2): + model = Joint_Monitoring() + # # # # TODO 需要运行编译一次,才能打印model.summary() + # model.build(input_shape=(batch_size, filter_num, dims)) + # model.summary() + history_loss = [] + history_val_loss = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + loss_value, accuracy_value = model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=True) + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy()) + k = 0 + z = z + 1 + k = k + 1 + val_loss, val_accuracy = model.get_val_loss(val_data=val_data, val_label1=val_label1, val_label2=val_label2, + is_first_time=True) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + # SaveBestH5Model(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy()) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + print('Training loss is :', loss_value.numpy()) + print('Validating loss is :', val_loss.numpy()) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def train_step_two(step_one_model, step_two_model, train_data, train_label1, train_label2): + # step_two_model = Joint_Monitoring() + # step_two_model.build(input_shape=(batch_size, time_stamp, feature_num)) + # step_two_model.summary() + history_loss = [] + history_val_loss = [] + history_accuracy = [] + learning_rate = 1e-3 + for epoch in range(EPOCH): + print() + print("EPOCH:", epoch, "/", EPOCH, ":") + train_data, train_label1, train_label2 = shuffle(train_data, train_label1, train_label2) + if epoch == 0: + train_data, train_label1, train_label2, val_data, val_label1, val_label2 = shuffle(train_data, train_label1, + train_label2, + is_split=True) + # print() + # print("EPOCH:", epoch, "/", EPOCH, ":") + # 用于让train知道，这是这个epoch中的第几次训练 + z = 0 + # 用于batch_size次再训练 + k = 1 + accuracy_num = 0 + for data_1, label_1, label_2 in zip(train_data, train_label1, train_label2): + size, _, _ = train_data.shape + data_1 = tf.expand_dims(data_1, axis=0) + label_1 = tf.expand_dims(label_1, axis=0) + label_2 = tf.expand_dims(label_2, axis=0) + if batch_size != 1: + if k % batch_size == 1: + data = data_1 + label1 = label_1 + label2 = label_2 + else: + data = tf.concat([data, data_1], axis=0) + label1 = tf.concat([label1, label_1], axis=0) + label2 = tf.concat([label2, label_2], axis=0) + else: + data = data_1 + label1 = label_1 + label2 = label_2 + + if k % batch_size == 0: + # label = tf.expand_dims(label, axis=-1) + output1, output2, output3, _ = step_one_model.call(inputs=data, is_first_time=True) + loss_value, accuracy_value = step_two_model.train(input_tensor=data, label1=label1, label2=label2, + learning_rate=learning_rate, + is_first_time=False, pred_3=output1, pred_4=output2, + pred_5=output3) + accuracy_num += accuracy_value + print(z * batch_size, "/", size, ":===============>", "loss:", loss_value.numpy(), "| accuracy:", + accuracy_num / ((z + 1) * batch_size)) + k = 0 + z = z + 1 + k = k + 1 + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=val_data, val_label1=val_label1, + val_label2=val_label2, + is_first_time=False, step_one_model=step_one_model) + SaveBestModelByAccuracy(model=step_two_model, save_name=save_step_two_name, history_accuracy=history_accuracy, + accuracy_value=val_accuracy) + history_val_loss.append(val_loss) + history_loss.append(loss_value.numpy()) + history_accuracy.append(val_accuracy) + print('Training loss is : {0} | Training accuracy is : {1}'.format(loss_value.numpy(), + accuracy_num / ((z + 1) * batch_size))) + print('Validating loss is : {0} | Validating accuracy is : {1}'.format(val_loss.numpy(), val_accuracy)) + if IsStopTraining(history_loss=history_val_loss, patience=7): + break + if Is_Reduce_learning_rate(history_loss=history_val_loss, patience=3): + if learning_rate >= 1e-4: + learning_rate = learning_rate * 0.1 + pass + + +def test(step_one_model, step_two_model, test_data, test_label1, test_label2): + history_loss = [] + history_val_loss = [] + + val_loss, val_accuracy = step_two_model.get_val_loss(val_data=test_data, val_label1=test_label1, + val_label2=test_label2, + is_first_time=False, step_one_model=step_one_model) + + history_val_loss.append(val_loss) + print("val_accuracy:", val_accuracy) + print("val_loss:", val_loss) + + +def showResult(step_two_model: Joint_Monitoring, test_data, isPlot: bool = False): + # 获取模型的所有参数的个数 + # step_two_model.count_params() + total_result = [] + size, length, dims = test_data.shape + for epoch in range(0, size - batch_size + 1, batch_size): + each_test_data = test_data[epoch:epoch + batch_size, :, :] + _, _, _, output4 = step_two_model.call(each_test_data, is_first_time=False) + total_result.append(output4) + total_result = np.reshape(total_result, [total_result.__len__(), -1]) + total_result = np.reshape(total_result, [-1, ]) + if isPlot: + plt.scatter(list(range(total_result.shape[0])), total_result, c='black', s=10) + # 画出 y=1 这条水平线 + plt.axhline(0.5, c='red', label='Failure threshold') + # 箭头指向上面的水平线 + # plt.arrow(35000, 0.9, 33000, 0.75, head_width=0.02, head_length=0.1, shape="full", fc='red', ec='red', + # alpha=0.9, overhang=0.5) + # plt.text(35000, 0.9, "Truth Fault", fontsize=10, color='black', verticalalignment='top') + plt.axvline(test_data.shape[0] * 2 / 3, c='blue', ls='-.') + plt.xlabel("time") + plt.ylabel("confience") + plt.text(total_result.shape[0] * 4 / 5, 0.6, "Fault", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10}) + plt.text(total_result.shape[0] * 1 / 3, 0.4, "Norm", fontsize=10, color='black', verticalalignment='top', + horizontalalignment='center', + bbox={'facecolor': 'grey', + 'pad': 10}) + plt.grid() + # plt.ylim(0, 1) + # plt.xlim(-50, 1300) + # plt.legend("", loc='upper left') + plt.show() + return total_result + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data1 = [] + predicted_data2 = [] + predicted_data3 = [] + size, length, dims = data.shape + for epoch in range(0, size, batch_size): + each_test_data = data[epoch:epoch + batch_size, :, :] + output1, output2, output3, _ = new_model.call(inputs=each_test_data, is_first_time=True) + if epoch == 0: + predicted_data1 = output1 + predicted_data2 = output2 + predicted_data3 = output3 + else: + predicted_data1 = np.concatenate([predicted_data1, output1], axis=0) + predicted_data2 = np.concatenate([predicted_data2, output2], axis=0) + predicted_data3 = np.concatenate([predicted_data3, output3], axis=0) + + predicted_data1 = np.reshape(predicted_data1, [-1, 10]) + predicted_data2 = np.reshape(predicted_data2, [-1, 10]) + predicted_data3 = np.reshape(predicted_data3, [-1, 10]) + predict_data = 0 + + predict_data = predicted_data1 + mseList = [] + meanList = [] + maxList = [] + + for i in range(1, 4): + print("i:", i) + if i == 1: + predict_data = predicted_data1 + elif i == 2: + predict_data = predicted_data2 + elif i == 3: + predict_data = predicted_data3 + temp = np.abs(predict_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predict_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predict_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + + print("mse.shape:", mse.shape) + # mse=np.mean((predicted_data-label)**2,axis=1) + # print("mse", mse) + mseList.append(mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max.shape:", max.shape) + # min = mean-3*std + max = np.broadcast_to(max, shape=[dims, ]) + # min = np.broadcast_to(min,shape=[dims,]) + mean = np.broadcast_to(mean, shape=[dims, ]) + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + maxList.append(max) + meanList.append(mean) + else: + if isPlot: + plt.figure(random.randint(1, 100)) + plt.plot(mse) + # plt.plot(min) + plt.show() + + return mseList, meanList, maxList + # pass + + +# healthy_data是健康数据,用于确定阈值，all_data是完整的数据,用于模型出结果 +def getResult(model: tf.keras.Model, healthy_data, healthy_label, unhealthy_data, unhealthy_label, isPlot: bool = False, + isSave: bool = False, predictI: int = 1): + # TODO 计算MSE确定阈值 + # plt.ion() + mseList, meanList, maxList = get_MSE(healthy_data, healthy_label, model) + mse1List, _, _ = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + for mse, mean, max, mse1,j in zip(mseList, meanList, maxList, mse1List,range(3)): + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + for i in range(total): + if (mse[i] > max[i]): + faultNum += 1 + faultList.append(mse[i]) + + fault_rate = faultNum / total + print("误报率:", fault_rate) + + # 漏报率计算 + missNum = 0 + missList = [] + all, = mse1.shape + for i in range(all): + if (mse1[i] < max[0]): + missNum += 1 + missList.append(mse1[i]) + + miss_rate = missNum / all + print("漏报率:", miss_rate) + + # 总体图 + print("mse:", mse) + print("mse1:", mse1) + print("============================================") + total_mse = np.concatenate([mse, mse1], axis=0) + total_max = np.broadcast_to(max[0], shape=[total_mse.shape[0], ]) + # min = np.broadcast_to(min,shape=[dims,]) + total_mean = np.broadcast_to(mean[0], shape=[total_mse.shape[0], ]) + + if isSave: + save_mse_name1=save_mse_name[:-4]+"_predict"+str(j+1)+".csv" + save_max_name1=save_max_name[:-4]+"_predict"+str(j+1)+".csv" + + np.savetxt(save_mse_name1,total_mse, delimiter=',') + np.savetxt(save_max_name1,total_max, delimiter=',') + + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +if __name__ == '__main__': + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + #### TODO 第一步训练 + # 单次测试 + # train_step_one(train_data=train_data_healthy[:32, :, :], train_label1=train_label1_healthy[:32, :],train_label2=train_label2_healthy[:32, ]) + train_step_one(train_data=train_data_healthy, train_label1=train_label1_healthy, train_label2=train_label2_healthy) + + + # 导入第一步已经训练好的模型,一个继续训练，一个只输出结果 + step_one_model = Joint_Monitoring() + step_one_model.load_weights(save_name) + # + # step_two_model = Joint_Monitoring() + # step_two_model.load_weights(save_name) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + all_data, _, _ = get_training_data_overlapping( + total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + + ##出结果单次测试 + # getResult(step_one_model, + # healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, + # :], + # healthy_label=train_label1_healthy[ + # healthy_size - 2 * unhealthy_size:healthy_size - 2 * unhealthy_size + 200, :], + # unhealthy_data=train_data_unhealthy[:200, :], unhealthy_label=train_label1_unhealthy[:200, :],isSave=True) + + getResult(step_one_model, healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + unhealthy_data=train_data_unhealthy, unhealthy_label=train_label1_unhealthy,isSave=True) + + ###TODO 展示全部的结果 + # all_data, _, _ = get_training_data_overlapping( + # total_data[healthy_size - 2 * unhealthy_size:unhealthy_date, :], is_Healthy=True) + # all_data = np.concatenate([]) + # 单次测试 + # showResult(step_two_model, test_data=all_data[:32], isPlot=True) + # showResult(step_two_model, test_data=all_data, isPlot=True) + + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/resnet_18.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/resnet_18.py new file mode 100644 index 0000000..8e196f1 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/resnet_18.py @@ -0,0 +1,310 @@ +import tensorflow as tf + +import numpy as np +from matplotlib import pyplot as plt +from sklearn.manifold import TSNE +from sklearn.metrics import confusion_matrix +import seaborn as sns +from sklearn.model_selection import train_test_split +from condition_monitoring.data_deal import loadData +from keras.callbacks import EarlyStopping + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 16 +learning_rate = 0.001 +EPOCH = 101 +model_name = "ResNet" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./model//{0}.h5".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_step_two_name = "./model/two_weight/{0}_weight_epoch6_99899_9996/weight".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +save_mse_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_result.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/RNet_C/{0}_timestamp{1}_feature{2}_max.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) + +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_total_zero.csv" + +''' +文件说明：jb4q_8_delete_total_zero.csv是删除了只删除了全是0的列的文件 +文件从0:415548行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:14:00) +从415549:432153行均是异常值(2019/9/18 11:21:01 - 2021/1/18 00:00:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 415548 +# 最后异常的时间点 +unhealthy_date = 432153 +# 异常容忍程度 +unhealthy_patience = 5 + +# 画图相关设置 +font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 10} # 设置坐标标签的字体大小，字体 + + +# train_data = np.load("../../../data/train_data.npy") +# train_label = np.load("../../../data/train_label.npy") +# test_data = np.load("../../../data/test_data.npy") +# test_label = np.load("../../../data/test_label.npy") + + +# CIFAR_100_data = tf.keras.datasets.cifar100 +# (train_data, train_label), (test_data, test_label) = CIFAR_100_data.load_data() +# train_data=np.array(train_data) +# train_label=np.array(train_label) +# print(train_data.shape) +# print(train_label.shape) +# print(train_data) +# print(test_data) +# +# +# 重叠采样 +def get_training_data_overlapping(data, time_stamp: int = time_stamp, is_Healthy: bool = True): + rows, cols = data.shape + train_data = np.empty(shape=[rows - time_stamp - 1, time_stamp, cols]) + train_label = np.empty(shape=[rows - time_stamp - 1, cols]) + for i in range(rows): + if i + time_stamp >= rows: + break + if i + time_stamp < rows - 1: + train_data[i] = data[i:i + time_stamp] + train_label[i] = data[i + time_stamp] + + print("重叠采样以后：") + print("data:", train_data) # (300334,120,10) + print("label:", train_label) # (300334,10) + + if is_Healthy: + train_label2 = np.ones(shape=[train_label.shape[0]]) + else: + train_label2 = np.zeros(shape=[train_label.shape[0]]) + + print("label2:", train_label2) + + return train_data, train_label, train_label2 + + +def shuffle(train_data, train_label1, train_label2, is_split: bool = False, split_size: float = 0.2): + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(train_data, + train_label1, + train_label2, + test_size=split_size, + shuffle=True, + random_state=100) + if is_split: + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + train_data = np.concatenate([train_data, test_data], axis=0) + train_label1 = np.concatenate([train_label1, test_label1], axis=0) + train_label2 = np.concatenate([train_label2, test_label2], axis=0) + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2 + pass + + +def split_test_data(healthy_data, healthy_label1, healthy_label2, unhealthy_data, unhealthy_label1, unhealthy_label2, + split_size: float = 0.2, shuffle: bool = True): + data = np.concatenate([healthy_data, unhealthy_data], axis=0) + label1 = np.concatenate([healthy_label1, unhealthy_label1], axis=0) + label2 = np.concatenate([healthy_label2, unhealthy_label2], axis=0) + (train_data, test_data, train_label1, test_label1, train_label2, test_label2) = train_test_split(data, + label1, + label2, + test_size=split_size, + shuffle=shuffle, + random_state=100) + + # print(train_data.shape) + # print(train_label1.shape) + # print(train_label2.shape) + # print(train_data.shape) + + return train_data, train_label1, train_label2, test_data, test_label1, test_label2 + + pass + + +# 归一化 +def normalization(data): + rows, cols = data.shape + print("归一化之前:", data) + print(data.shape) + print("======================") + + # 归一化 + max = np.max(data, axis=0) + max = np.broadcast_to(max, [rows, cols]) + min = np.min(data, axis=0) + min = np.broadcast_to(min, [rows, cols]) + + data = (data - min) / (max - min) + print("归一化之后:", data) + print(data.shape) + + return data + + +def identity_block(input_tensor, out_dim): + con1 = tf.keras.layers.Conv1D(filters=out_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + input_tensor) + bhn1 = tf.keras.layers.BatchNormalization()(con1) + + # con2 = tf.keras.layers.Conv1D(filters=out_dim // 4, kernel_size=3, padding='SAME', activation=tf.nn.relu)(bhn1) + # bhn2 = tf.keras.layers.BatchNormalization()(con2) + + con3 = tf.keras.layers.Conv1D(filters=out_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)(bhn1) + + out = tf.keras.layers.Add()([input_tensor, con3]) + out = tf.nn.relu(out) + + return out + + +def resnet_Model(): + inputs = tf.keras.Input(shape=[120, 10]) + conv1 = tf.keras.layers.Conv1D(filters=20, kernel_size=3, padding='SAME', activation=tf.nn.relu)(inputs) + '''第一层''' + output_dim = 10 + identity_1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + conv1) + identity_1 = tf.keras.layers.BatchNormalization()(identity_1) + for _ in range(2): + identity_1 = identity_block(identity_1, output_dim) + '''第二层''' + output_dim = 20 + identity_2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_1) + identity_2 = tf.keras.layers.BatchNormalization()(identity_2) + for _ in range(2): + identity_2 = identity_block(identity_2, output_dim) + '''第三层''' + output_dim = 20 + identity_3 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_2) + identity_3 = tf.keras.layers.BatchNormalization()(identity_3) + for _ in range(2): + identity_3 = identity_block(identity_3, output_dim) + '''第四层''' + output_dim = 40 + identity_4 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_3) + identity_4 = tf.keras.layers.BatchNormalization()(identity_4) + for _ in range(2): + identity_4 = identity_block(identity_4, output_dim) + flatten = tf.keras.layers.GlobalAvgPool1D()(identity_4) + dropout = tf.keras.layers.Dropout(0.217)(flatten) + + dense = tf.keras.layers.Dense(128, activation=tf.nn.relu)(dropout) + dense = tf.keras.layers.BatchNormalization(name="bn_last")(dense) + dense = tf.keras.layers.Dense(2, activation=tf.nn.sigmoid)(dense) + model = tf.keras.Model(inputs=inputs, outputs=dense) + return model + + +if __name__ == '__main__': + # # 数据读入 + # + total_data = loadData.execute(N=feature_num, file_name=file_name) + total_data = normalization(data=total_data) + train_data_healthy, train_label1_healthy, train_label2_healthy = get_training_data_overlapping( + total_data[:healthy_date, :], is_Healthy=True) + train_data_unhealthy, train_label1_unhealthy, train_label2_unhealthy = get_training_data_overlapping( + total_data[healthy_date - time_stamp + unhealthy_patience:unhealthy_date, :], + is_Healthy=False) + + healthy_size, _, _ = train_data_healthy.shape + unhealthy_size, _, _ = train_data_unhealthy.shape + train_data, train_label1, train_label2, test_data, test_label1, test_label2 = split_test_data( + healthy_data=train_data_healthy[healthy_size - 2 * unhealthy_size:, :, :], + healthy_label1=train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], + healthy_label2=train_label2_healthy[healthy_size - 2 * unhealthy_size:, ], unhealthy_data=train_data_unhealthy, + unhealthy_label1=train_label1_unhealthy, unhealthy_label2=train_label2_unhealthy) + + train_label = train_label2 + test_label = test_label2 + + model = resnet_Model() + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.binary_crossentropy, + metrics=['acc']) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=5, mode='min', verbose=1) + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=1, + save_best_only=True, + mode='min', + period=1) + + history = model.fit(train_data, train_label, epochs=20, batch_size=32, validation_data=(test_data, test_label), + callbacks=[checkpoint, early_stop]) + model.save("./model/ResNet.h5") + model = tf.keras.models.load_model("../model/ResNet_model.h5") + + # 结果展示 + trained_data = tf.keras.models.Model(inputs=model.input, outputs=model.get_layer('bn_last').output).predict( + train_data) + predict_label = model.predict(test_data) + predict_label_max = np.argmax(predict_label, axis=1) + predict_label = np.expand_dims(predict_label_max, axis=1) + + confusion_matrix = confusion_matrix(test_label, predict_label) + + tsne = TSNE(n_components=3, verbose=2, perplexity=30, n_iter=5000).fit_transform(trained_data) + print("tsne[:,0]", tsne[:, 0]) + print("tsne[:,1]", tsne[:, 1]) + print("tsne[:,2]", tsne[:, 2]) + x, y, z = tsne[:, 0], tsne[:, 1], tsne[:, 2] + x = (x - np.min(x)) / (np.max(x) - np.min(x)) + y = (y - np.min(y)) / (np.max(y) - np.min(y)) + z = (z - np.min(z)) / (np.max(z) - np.min(z)) + + fig1 = plt.figure() + ax1 = fig1.add_subplot(projection='3d') + ax1.scatter3D(x, y, z, c=train_label, cmap=plt.cm.get_cmap("jet", 10)) + + fig2 = plt.figure() + ax2 = fig2.add_subplot() + sns.heatmap(confusion_matrix, annot=True, fmt="d", cmap='Blues') + plt.ylabel('Actual label') + plt.xlabel('Predicted label') + + # fig3 = plt.figure() + # ax3 = fig3.add_subplot() + # plt.plot(history.epoch, history.history.get('acc'), label='acc') + # plt.plot(history.epoch, history.history.get('val_acc'), label='val_acc') + # + # fig4 = plt.figure() + # ax4 = fig3.add_subplot() + # plt.plot(history.epoch, history.history.get('loss'), label='loss') + # plt.plot(history.epoch, history.history.get('val_loss'), label='val_loss') + plt.legend() + plt.show() + + score = model.evaluate(test_data, test_label) + print('score:', score) diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-E.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/svm.py similarity index 52% rename from TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-E.py rename to TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/svm.py index 18442fc..766a9a5 100644 --- a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet-E.py +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/svm.py @@ -4,7 +4,12 @@ ''' @Author : dingjiawen -@Date : 2022/10/11 18:53 +@Date : 2022/10/19 14:34 @Usage : -@Desc : -''' \ No newline at end of file +@Desc : SVM +''' + +import sklearn.svm as svm + + + diff --git a/TensorFlow_eaxmple/Model_train_test/model/Dynamic_channelAttention/SE_channelAttention.py b/TensorFlow_eaxmple/Model_train_test/model/Dynamic_channelAttention/SE_channelAttention.py new file mode 100644 index 0000000..63f286c --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Dynamic_channelAttention/SE_channelAttention.py @@ -0,0 +1,129 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/12 17:48 +@Usage : 经典SE通道注意力 +@Desc : +''' + +import tensorflow as tf +import tensorflow.keras +from tensorflow.keras import * +import tensorflow.keras.layers as layers +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D + +import keras.backend as K + + +class Between_0_1(tf.keras.constraints.Constraint): + def __call__(self, w): + # 调用父类__init__()方法 + super(Between_0_1, self).__init__() + return K.clip(w, 0, 1) + + +class SEChannelAttention(layers.Layer): + + def __init__(self): + # 调用父类__init__()方法 + super(SEChannelAttention, self).__init__() + self.DWC = DepthwiseConv1D(kernel_size=1, padding='SAME') + # self.DWC = DepthwiseConv1D(kernel_size=1, padding='causal',dilation_rate=4,data_format='channels_last') + + def build(self, input_shape): + if len(input_shape) != 3: + raise ValueError('Inputs to `DynamicChannelAttention` should have rank 3. ' + 'Received input shape:', str(input_shape)) + + print(input_shape) + # GAP + self.GAP = tf.keras.layers.GlobalAvgPool1D() + self.c1 = tf.keras.layers.Conv1D(filters=input_shape[2], kernel_size=1, padding='SAME') + # s1 = tf.nn.sigmoid(c1) + + # GMP + self.GMP = tf.keras.layers.GlobalMaxPool1D() + self.c2 = tf.keras.layers.Conv1D(filters=input_shape[2], kernel_size=1, padding='SAME') + # s2 = tf.nn.sigmoid(c2) + + # weight + self.weight_kernel = self.add_weight( + shape=(1, input_shape[2]), + initializer='glorot_uniform', + name='weight_kernel') + + + + def call(self, inputs, **kwargs): + batch_size, length, channel = inputs.shape + DWC1 = self.DWC(inputs) + + # GAP + GAP = self.GAP(DWC1) + GAP = tf.expand_dims(GAP, axis=1) + c1 = self.c1(GAP) + # c1 = tf.keras.layers.BatchNormalization()(c1) + # s1 = tf.nn.sigmoid(c1) + + # # GMP + # GMP = self.GMP(DWC1) + # GMP = tf.expand_dims(GMP, axis=1) + # c2 = self.c2(GMP) + # c2 = tf.keras.layers.BatchNormalization()(c2) + # s2 = tf.nn.sigmoid(c2) + + # print(self.weight_kernel) + + weight_kernel = tf.broadcast_to(self.weight_kernel, shape=[length, channel]) + weight_kernel = tf.broadcast_to(weight_kernel, shape=[batch_size, length, channel]) + s1 = tf.broadcast_to(c1, shape=[batch_size, length, channel]) + # s2 = tf.broadcast_to(s2, shape=[batch_size, length, channel]) + + output = weight_kernel * s1 * inputs + return output + + + +class DynamicPooling(layers.Layer): + + def __init__(self, pool_size=2): + # 调用父类__init__()方法 + super(DynamicPooling, self).__init__() + self.pool_size = pool_size + pass + + def build(self, input_shape): + if len(input_shape) != 3: + raise ValueError('Inputs to `DynamicChannelAttention` should have rank 3. ' + 'Received input shape:', str(input_shape)) + # GAP + self.AP = tf.keras.layers.AveragePooling1D(pool_size=self.pool_size) + + # GMP + self.MP = tf.keras.layers.MaxPool1D(pool_size=self.pool_size) + + # weight + self.weight_kernel = self.add_weight( + shape=(int(input_shape[1] / self.pool_size), input_shape[2]), + initializer='glorot_uniform', + name='weight_kernel', + constraint=Between_0_1()) + + def call(self, inputs, **kwargs): + batch_size, length, channel = inputs.shape + + # GAP + GAP = self.AP(inputs) + + # GMP + GMP = self.MP(inputs) + + weight_kernel = tf.broadcast_to(self.weight_kernel, shape=GMP.shape) + + output = tf.add(weight_kernel * GAP, (tf.ones_like(weight_kernel) - weight_kernel) * GMP) + return output diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_3.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_3.py new file mode 100644 index 0000000..80ed640 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_3.py @@ -0,0 +1,457 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + concat3=output1 + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + concat3 = output1 + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_34.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_34.py new file mode 100644 index 0000000..fc76c58 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_34.py @@ -0,0 +1,455 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output2], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output2], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_35.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_35.py new file mode 100644 index 0000000..f738d2b --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_35.py @@ -0,0 +1,455 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_4.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_4.py new file mode 100644 index 0000000..2a7323b --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_4.py @@ -0,0 +1,457 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + concat3 = output2 + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + concat3 = output2 + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_45.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_45.py new file mode 100644 index 0000000..60ca054 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_45.py @@ -0,0 +1,455 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_5.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_5.py new file mode 100644 index 0000000..7625d36 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_5.py @@ -0,0 +1,457 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:DCAU只分类，不预测 +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = DynamicChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = DynamicChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = DynamicChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + # step three + # 分类器 + self.d4 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.d5 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + # tf.nn.softmax + self.output4 = tf.keras.layers.Dense(1, activation=tf.nn.sigmoid) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + concat3 = output3 + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + # concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + concat3 = output3 + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss= 0 + # MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + # MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + # MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_S.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_S.py new file mode 100644 index 0000000..c55dd12 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet_S.py @@ -0,0 +1,447 @@ +# _*_ coding: UTF-8 _*_ + + +''' +@Author : dingjiawen +@Date : 2022/7/14 9:40 +@Usage : 联合监测模型 +@Desc : RNet:LCAU +''' + +import tensorflow as tf +import tensorflow.keras as keras +from tensorflow.keras import * +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from model.DepthwiseCon1D.DepthwiseConv1D import DepthwiseConv1D +from model.Dynamic_channelAttention.SE_channelAttention import SEChannelAttention, DynamicPooling +from condition_monitoring.data_deal import loadData +from model.LossFunction.smooth_L1_Loss import SmoothL1Loss + + +class Joint_Monitoring(keras.Model): + + def __init__(self, conv_filter=20): + # 调用父类__init__()方法 + super(Joint_Monitoring, self).__init__() + # step one + self.RepDCBlock1 = RevConvBlock(num=3, kernel_size=5) + self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample1 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU2 = SEChannelAttention() + self.RepDCBlock2 = RevConvBlock(num=3, kernel_size=3) + self.conv2 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=1, strides=2, padding='SAME') + self.upsample2 = tf.keras.layers.UpSampling1D(size=2) + + self.DACU3 = SEChannelAttention() + self.RepDCBlock3 = RevConvBlock(num=3, kernel_size=3) + self.p1 = DynamicPooling(pool_size=2) + self.conv3 = tf.keras.layers.Conv1D(filters=2 * conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.DACU4 = SEChannelAttention() + self.RepDCBlock4 = RevConvBlock(num=3, kernel_size=3) + self.p2 = DynamicPooling(pool_size=4) + self.conv4 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=3, strides=2, padding='SAME') + + self.RepDCBlock5 = RevConvBlock(num=3, kernel_size=3) + self.p3 = DynamicPooling(pool_size=2) + + # step two + # 重现原数据 + self.GRU1 = tf.keras.layers.GRU(128, return_sequences=False) + self.d1 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output1 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU2 = tf.keras.layers.GRU(128, return_sequences=False) + self.d2 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output2 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + self.GRU3 = tf.keras.layers.GRU(128, return_sequences=False) + self.d3 = tf.keras.layers.Dense(300, activation=tf.nn.leaky_relu) + self.output3 = tf.keras.layers.Dense(10, activation=tf.nn.leaky_relu) + + + # loss + self.train_loss = [] + + def call(self, inputs, training=None, mask=None, is_first_time: bool = True): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + output1 = [] + output2 = [] + output3 = [] + output4 = [] + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + else: + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + return output1, output2, output3, output4 + + def get_loss(self, inputs_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + # step one + RepDCBlock1 = self.RepDCBlock1(inputs_tensor) + RepDCBlock1 = tf.keras.layers.BatchNormalization()(RepDCBlock1) + conv1 = self.conv1(RepDCBlock1) + conv1 = tf.nn.leaky_relu(conv1) + conv1 = tf.keras.layers.BatchNormalization()(conv1) + upsample1 = self.upsample1(conv1) + + DACU2 = self.DACU2(upsample1) + DACU2 = tf.keras.layers.BatchNormalization()(DACU2) + RepDCBlock2 = self.RepDCBlock2(DACU2) + RepDCBlock2 = tf.keras.layers.BatchNormalization()(RepDCBlock2) + conv2 = self.conv2(RepDCBlock2) + conv2 = tf.nn.leaky_relu(conv2) + conv2 = tf.keras.layers.BatchNormalization()(conv2) + upsample2 = self.upsample2(conv2) + + DACU3 = self.DACU3(upsample2) + DACU3 = tf.keras.layers.BatchNormalization()(DACU3) + RepDCBlock3 = self.RepDCBlock3(DACU3) + RepDCBlock3 = tf.keras.layers.BatchNormalization()(RepDCBlock3) + conv3 = self.conv3(RepDCBlock3) + conv3 = tf.nn.leaky_relu(conv3) + conv3 = tf.keras.layers.BatchNormalization()(conv3) + + concat1 = tf.concat([conv2, conv3], axis=1) + + DACU4 = self.DACU4(concat1) + DACU4 = tf.keras.layers.BatchNormalization()(DACU4) + RepDCBlock4 = self.RepDCBlock4(DACU4) + RepDCBlock4 = tf.keras.layers.BatchNormalization()(RepDCBlock4) + conv4 = self.conv4(RepDCBlock4) + conv4 = tf.nn.leaky_relu(conv4) + conv4 = tf.keras.layers.BatchNormalization()(conv4) + + concat2 = tf.concat([conv1, conv4], axis=1) + + RepDCBlock5 = self.RepDCBlock5(concat2) + RepDCBlock5 = tf.keras.layers.BatchNormalization()(RepDCBlock5) + + if is_first_time: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # reduce_mean降维计算均值 + MSE_loss1 = SmoothL1Loss()(y_true=label1, y_pred=output1) + MSE_loss2 = SmoothL1Loss()(y_true=label1, y_pred=output2) + MSE_loss3 = SmoothL1Loss()(y_true=label1, y_pred=output3) + # MSE_loss1 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output1)) + # MSE_loss2 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output2)) + # MSE_loss3 = tf.reduce_mean(tf.keras.losses.mse(y_true=label1, y_pred=output3)) + + print("MSE_loss1:", MSE_loss1.numpy()) + print("MSE_loss2:", MSE_loss2.numpy()) + print("MSE_loss3:", MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 + Accuracy_num = 0 + + else: + # step two + # 重现原数据 + # 接block3 + GRU1 = self.GRU1(RepDCBlock3) + GRU1 = tf.keras.layers.BatchNormalization()(GRU1) + d1 = self.d1(GRU1) + # tf.nn.softmax + output1 = self.output1(d1) + # 接block4 + GRU2 = self.GRU2(RepDCBlock4) + GRU2 = tf.keras.layers.BatchNormalization()(GRU2) + d2 = self.d2(GRU2) + # tf.nn.softmax + output2 = self.output2(d2) + # 接block5 + GRU3 = self.GRU3(RepDCBlock5) + GRU3 = tf.keras.layers.BatchNormalization()(GRU3) + d3 = self.d3(GRU3) + # tf.nn.softmax + output3 = self.output3(d3) + + # 多尺度动态池化 + # p1 = self.p1(output1) + # B, _, _ = p1.shape + # f1 = tf.reshape(p1, shape=[B, -1]) + # p2 = self.p2(output2) + # f2 = tf.reshape(p2, shape=[B, -1]) + # p3 = self.p3(output3) + # f3 = tf.reshape(p3, shape=[B, -1]) + # step three + # 分类器 + concat3 = tf.concat([output1, output2, output3], axis=1) + # dropout = tf.keras.layers.Dropout(0.25)(concat3) + d4 = self.d4(concat3) + d5 = self.d5(d4) + # d4 = tf.keras.layers.BatchNormalization()(d4) + output4 = self.output4(d5) + + # reduce_mean降维计算均值 + MSE_loss = SmoothL1Loss()(y_true=pred_3, y_pred=output1) + MSE_loss += SmoothL1Loss()(y_true=pred_4, y_pred=output2) + MSE_loss += SmoothL1Loss()(y_true=pred_5, y_pred=output3) + Cross_Entropy_loss = tf.reduce_mean( + tf.losses.binary_crossentropy(y_true=label2, y_pred=output4, from_logits=True)) + + print("MSE_loss:", MSE_loss.numpy()) + print("Cross_Entropy_loss:", Cross_Entropy_loss.numpy()) + Accuracy_num = self.get_Accuracy(label=label2, output=output4) + loss = MSE_loss + Cross_Entropy_loss + return loss, Accuracy_num + + def get_Accuracy(self, output, label): + + predict_label = tf.round(output) + label = tf.cast(label, dtype=tf.float32) + + t = np.array(label - predict_label) + + b = t[t[:] == 0] + + return b.__len__() + + def get_grad(self, input_tensor, label1=None, label2=None, is_first_time: bool = True, pred_3=None, pred_4=None, + pred_5=None): + with tf.GradientTape() as tape: + # todo 原本tape只会监控由tf.Variable创建的trainable=True属性 + # tape.watch(self.variables) + L, Accuracy_num = self.get_loss(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + # 保存一下loss，用于输出 + self.train_loss = L + g = tape.gradient(L, self.variables) + return g, Accuracy_num + + def train(self, input_tensor, label1=None, label2=None, learning_rate=1e-3, is_first_time: bool = True, pred_3=None, + pred_4=None, pred_5=None): + g, Accuracy_num = self.get_grad(input_tensor, label1=label1, label2=label2, is_first_time=is_first_time, + pred_3=pred_3, + pred_4=pred_4, pred_5=pred_5) + optimizers.Adam(learning_rate).apply_gradients(zip(g, self.variables)) + return self.train_loss, Accuracy_num + + # 暂时只支持batch_size等于1,不然要传z比较麻烦 + def get_val_loss(self, val_data, val_label1, val_label2, batch_size=16, is_first_time: bool = True, + step_one_model=None): + val_loss = [] + accuracy_num = 0 + output1 = 0 + output2 = 0 + output3 = 0 + z = 1 + size, length, dims = val_data.shape + if batch_size == None: + batch_size = self.batch_size + for epoch in range(0, size - batch_size, batch_size): + each_val_data = val_data[epoch:epoch + batch_size, :, :] + each_val_label1 = val_label1[epoch:epoch + batch_size, :] + each_val_label2 = val_label2[epoch:epoch + batch_size, ] + # each_val_data = tf.expand_dims(each_val_data, axis=0) + # each_val_query = tf.expand_dims(each_val_query, axis=0) + # each_val_label = tf.expand_dims(each_val_label, axis=0) + if not is_first_time: + output1, output2, output3, _ = step_one_model.call(inputs=each_val_data, is_first_time=True) + + each_loss, each_accuracy_num = self.get_loss(each_val_data, each_val_label1, each_val_label2, + is_first_time=is_first_time, + pred_3=output1, pred_4=output2, pred_5=output3) + accuracy_num += each_accuracy_num + val_loss.append(each_loss) + z += 1 + + val_accuracy = accuracy_num / ((z-1) * batch_size) + val_total_loss = tf.reduce_mean(val_loss) + return val_total_loss, val_accuracy + + +class RevConv(keras.layers.Layer): + + def __init__(self, kernel_size=3): + # 调用父类__init__()方法 + super(RevConv, self).__init__() + self.kernel_size = kernel_size + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size + } + ) + base_config = super(RevConv, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # print(input_shape) + _, _, output_dim = input_shape[0], input_shape[1], input_shape[2] + self.conv1 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=self.kernel_size, strides=1, + padding='causal', + dilation_rate=4) + + self.conv2 = tf.keras.layers.Conv1D(filters=output_dim, kernel_size=1, strides=1, padding='causal', + dilation_rate=4) + # self.b2 = tf.keras.layers.BatchNormalization() + + # self.b3 = tf.keras.layers.BatchNormalization() + + # out = tf.keras.layers.Add()([b1, b2, b3]) + # out = tf.nn.relu(out) + + def call(self, inputs, **kwargs): + conv1 = self.conv1(inputs) + b1 = tf.keras.layers.BatchNormalization()(conv1) + b1 = tf.nn.leaky_relu(b1) + # b1 = self.b1 + + conv2 = self.conv2(inputs) + b2 = tf.keras.layers.BatchNormalization()(conv2) + b2 = tf.nn.leaky_relu(b2) + + b3 = tf.keras.layers.BatchNormalization()(inputs) + + out = tf.keras.layers.Add()([b1, b2, b3]) + out = tf.nn.relu(out) + + return out + + +class RevConvBlock(keras.layers.Layer): + + def __init__(self, num: int = 3, kernel_size=3): + # 调用父类__init__()方法 + super(RevConvBlock, self).__init__() + self.num = num + self.kernel_size = kernel_size + self.L = [] + for i in range(num): + RepVGG = RevConv(kernel_size=kernel_size) + self.L.append(RepVGG) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'kernel_size': self.kernel_size, + 'num': self.num + } + ) + base_config = super(RevConvBlock, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, **kwargs): + for i in range(self.num): + inputs = self.L[i](inputs) + return inputs diff --git a/TensorFlow_eaxmple/Model_train_test/model/ResNet/train/resnet_18.py b/TensorFlow_eaxmple/Model_train_test/model/ResNet/train/resnet_18.py new file mode 100644 index 0000000..a45b891 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/ResNet/train/resnet_18.py @@ -0,0 +1,133 @@ +import tensorflow as tf + +import numpy as np +from matplotlib import pyplot as plt +from sklearn.manifold import TSNE +from sklearn.metrics import confusion_matrix +import seaborn as sns + +# 数据读入 +train_data = np.load("../../../data/train_data.npy") +train_label = np.load("../../../data/train_label.npy") +test_data = np.load("../../../data/test_data.npy") +test_label = np.load("../../../data/test_label.npy") + + +# CIFAR_100_data = tf.keras.datasets.cifar100 +# (train_data, train_label), (test_data, test_label) = CIFAR_100_data.load_data() +# train_data=np.array(train_data) +# train_label=np.array(train_label) +# print(train_data.shape) +# print(train_label.shape) +# print(train_data) +# print(test_data) +# +# +def identity_block(input_tensor, out_dim): + con1 = tf.keras.layers.Conv2D(filters=out_dim // 4, kernel_size=1, padding='SAME', activation=tf.nn.relu)( + input_tensor) + bhn1 = tf.keras.layers.BatchNormalization()(con1) + + con2 = tf.keras.layers.Conv2D(filters=out_dim // 4, kernel_size=3, padding='SAME', activation=tf.nn.relu)(bhn1) + bhn2 = tf.keras.layers.BatchNormalization()(con2) + + con3 = tf.keras.layers.Conv2D(filters=out_dim, kernel_size=1, padding='SAME', activation=tf.nn.relu)(bhn2) + + out = tf.keras.layers.Add()([input_tensor, con3]) + out = tf.nn.relu(out) + + return out + + +def resnet_Model(): + inputs = tf.keras.Input(shape=[80, 80, 9]) + conv1 = tf.keras.layers.Conv2D(filters=64, kernel_size=3, padding='SAME', activation=tf.nn.relu)(inputs) + '''第一层''' + output_dim = 64 + identity_1 = tf.keras.layers.Conv2D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + conv1) + identity_1 = tf.keras.layers.BatchNormalization()(identity_1) + for _ in range(3): + identity_1 = identity_block(identity_1, output_dim) + '''第二层''' + output_dim = 128 + identity_2 = tf.keras.layers.Conv2D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_1) + identity_2 = tf.keras.layers.BatchNormalization()(identity_2) + for _ in range(2): + identity_2 = identity_block(identity_2, output_dim) + '''第三层''' + output_dim = 256 + identity_3 = tf.keras.layers.Conv2D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_2) + identity_3 = tf.keras.layers.BatchNormalization()(identity_3) + for _ in range(3): + identity_3 = identity_block(identity_3, output_dim) + '''第四层''' + output_dim = 512 + identity_4 = tf.keras.layers.Conv2D(filters=output_dim, kernel_size=3, padding='SAME', activation=tf.nn.relu)( + identity_3) + identity_4 = tf.keras.layers.BatchNormalization()(identity_4) + for _ in range(3): + identity_4 = identity_block(identity_4, output_dim) + flatten = tf.keras.layers.Flatten()(identity_4) + dropout = tf.keras.layers.Dropout(0.217)(flatten) + + dense = tf.keras.layers.Dense(128, activation=tf.nn.relu,)(dropout) + dense = tf.keras.layers.BatchNormalization(name="bn_last")(dense) + dense = tf.keras.layers.Dense(9, activation=tf.nn.softmax)(dense) + model = tf.keras.Model(inputs=inputs, outputs=dense) + return model + + +if __name__ == '__main__': + model = resnet_Model() + model.compile(optimizer=tf.optimizers.SGD(1e-3,momentum=0.02), loss=tf.losses.sparse_categorical_crossentropy, + metrics=['acc']) + model.summary() + history = model.fit(train_data, train_label, epochs=10, batch_size=10,validation_data=(test_data, test_label)) + model.save("ResNet_model.h5") + # model=tf.keras.models.load_model("../model/ResNet_model.h5") + # + # trained_data = tf.keras.models.Model(inputs=model.input, outputs=model.get_layer('bn_last').output).predict( + # train_data) + # predict_label = model.predict(test_data) + # predict_label_max = np.argmax(predict_label, axis=1) + # predict_label = np.expand_dims(predict_label_max, axis=1) + # + # confusion_matrix = confusion_matrix(test_label, predict_label) + + tsne = TSNE(n_components=3, verbose=2, perplexity=30,n_iter=5000).fit_transform(trained_data) + print("tsne[:,0]", tsne[:, 0]) + print("tsne[:,1]", tsne[:, 1]) + print("tsne[:,2]", tsne[:, 2]) + x, y, z = tsne[:, 0], tsne[:, 1], tsne[:, 2] + x = (x - np.min(x)) / (np.max(x) - np.min(x)) + y = (y - np.min(y)) / (np.max(y) - np.min(y)) + z = (z - np.min(z)) / (np.max(z) - np.min(z)) + + fig1 = plt.figure() + ax1 = fig1.add_subplot(projection='3d') + ax1.scatter3D(x, y, z, c=train_label, cmap=plt.cm.get_cmap("jet", 10)) + + fig2 = plt.figure() + ax2 = fig2.add_subplot() + sns.heatmap(confusion_matrix, annot=True, fmt="d", cmap='Blues') + plt.ylabel('Actual label') + plt.xlabel('Predicted label') + + + # fig3 = plt.figure() + # ax3 = fig3.add_subplot() + # plt.plot(history.epoch, history.history.get('acc'), label='acc') + # plt.plot(history.epoch, history.history.get('val_acc'), label='val_acc') + # + # fig4 = plt.figure() + # ax4 = fig3.add_subplot() + # plt.plot(history.epoch, history.history.get('loss'), label='loss') + # plt.plot(history.epoch, history.history.get('val_loss'), label='val_loss') + plt.legend() + plt.show() + + score = model.evaluate(test_data, test_label) + print('score:', score)