diff --git a/Spider/线程相关/__init__.py b/Spider/chapter01_线程相关/__init__.py similarity index 100% rename from Spider/线程相关/__init__.py rename to Spider/chapter01_线程相关/__init__.py diff --git a/Spider/线程相关/多线程/MyThread.py b/Spider/chapter01_线程相关/多线程/MyThread.py similarity index 100% rename from Spider/线程相关/多线程/MyThread.py rename to Spider/chapter01_线程相关/多线程/MyThread.py diff --git a/Spider/线程相关/多线程/__init__.py b/Spider/chapter01_线程相关/多线程/__init__.py similarity index 100% rename from Spider/线程相关/多线程/__init__.py rename to Spider/chapter01_线程相关/多线程/__init__.py diff --git a/Spider/线程相关/多进程/Lock.py b/Spider/chapter01_线程相关/多进程/Lock.py similarity index 100% rename from Spider/线程相关/多进程/Lock.py rename to Spider/chapter01_线程相关/多进程/Lock.py diff --git a/Spider/线程相关/多进程/MyMutiProcess.py b/Spider/chapter01_线程相关/多进程/MyMutiProcess.py similarity index 100% rename from Spider/线程相关/多进程/MyMutiProcess.py rename to Spider/chapter01_线程相关/多进程/MyMutiProcess.py diff --git a/Spider/线程相关/多进程/Semaphore.py b/Spider/chapter01_线程相关/多进程/Semaphore.py similarity index 100% rename from Spider/线程相关/多进程/Semaphore.py rename to Spider/chapter01_线程相关/多进程/Semaphore.py diff --git a/Spider/线程相关/多进程/__init__.py b/Spider/chapter01_线程相关/多进程/__init__.py similarity index 100% rename from Spider/线程相关/多进程/__init__.py rename to Spider/chapter01_线程相关/多进程/__init__.py diff --git a/Spider/线程相关/多进程/pool.py b/Spider/chapter01_线程相关/多进程/pool.py similarity index 100% rename from Spider/线程相关/多进程/pool.py rename to Spider/chapter01_线程相关/多进程/pool.py diff --git a/Spider/chapter02_爬虫基本库/Practice.py b/Spider/chapter02_爬虫基本库/Practice.py new file mode 100644 index 0000000..86d3260 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/Practice.py @@ -0,0 +1,164 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 19:09 +@Usage : +@Desc : 一个基本的练习：爬取 https://ssr1.scrape.center 电影描述以及detail等 +''' + +import requests +import logging +import re +from urllib.parse import urljoin +from os import makedirs +from os.path import exists +from pyquery import PyQuery as pq +import json + +# 输出的日志级别 +logging.basicConfig(level=logging.INFO, + format='%(asctime)s - %(levelname)s: %(message)s') +BASE_URL = 'https://ssr1.scrape.center' +TOTAL_PAGE = 10 + +RESULTS_DIR = 'results' +exists(RESULTS_DIR) or makedirs(RESULTS_DIR) + + +def scrape_page(url): + """ + scrape page by url and return its html + :param url: page url + :return: html of page + """ + logging.info('scraping %s...', url) + try: + response = requests.get(url) + if response.status_code == 200: + return response.text + logging.error('get invalid status code %s while scraping %s', response.status_code, url) + except requests.RequestException: + logging.error('error occurred while scraping %s', url, exc_info=True) + + +def scrape_index(page): + index_url = f'{BASE_URL}/page/{page}' + return scrape_page(index_url) + + +# 从页面HTML中获取详情页坐标 +def parse_index(html): + pattern = '' + items = re.findall(pattern, html, re.S) + if not items: + return [] + for item in items: + detail_url = urljoin(BASE_URL, item) + logging.info('get detail url %s', detail_url) + yield detail_url + + +def scrape_detail(url): + return scrape_page(url) + + +def parse_detail(html): + cover_pattern = re.compile( + 'class="item.*?', re.S) + name_pattern = re.compile('(.*?)') + categories_pattern = re.compile( + '(.*?).*?', re.S) + published_at_pattern = re.compile('(\d{4}-\d{2}-\d{2})\s?上映') + drama_pattern = re.compile('.*?(.*?)

', re.S) + score_pattern = re.compile('(.*?)

', re.S) + + cover = re.search(cover_pattern, html).group( + 1).strip() if re.search(cover_pattern, html) else None + name = re.search(name_pattern, html).group( + 1).strip() if re.search(name_pattern, html) else None + categories = re.findall(categories_pattern, html) if re.findall( + categories_pattern, html) else [] + published_at = re.search(published_at_pattern, html).group( + 1) if re.search(published_at_pattern, html) else None + drama = re.search(drama_pattern, html).group( + 1).strip() if re.search(drama_pattern, html) else None + score = float(re.search(score_pattern, html).group(1).strip() + ) if re.search(score_pattern, html) else None + return { + 'cover': cover, + 'name': name, + 'categories': categories, + 'published_at': published_at, + 'drama': drama, + 'score': score + } + + +def parse_detailByPyQuery(html): + doc = pq(html) + cover = doc('img.cover').attr('src') + name = doc('a > h2').text() + categories = [item.text() for item in doc('.categories button span').items()] + published_at = doc('.info:contains(上映)').text() + published_at = re.search('(\d{4}-\d{2}-\d{2})', published_at).group(1) \ + if published_at and re.search('\d{4}-\d{2}-\d{2}', published_at) else None + drama = doc('.drama p').text() + score = doc('p.score').text() + score = float(score) if score else None + return { + 'cover': cover, + 'name': name, + 'categories': categories, + 'published_at': published_at, + 'drama': drama, + 'score': score + } + + +def save_data(data): + """ + save to json file + :param data: + :return: + """ + name = data.get('name') + data_path = f'{RESULTS_DIR}/{name}.json' + # ensure_ascii值为False,可以保证中文字符在文件内能以正常的正文文本呈现，而不是unicode + # indent为2,可以使json可以两行缩进 + json.dump(data, open(data_path, 'w', encoding='utf-8'), + ensure_ascii=False, indent=2) + + +def main(): + for page in range(1, TOTAL_PAGE): + index_html = scrape_index(page) + detail_urls = parse_index(index_html) + for detail_url in detail_urls: + data = parse_detail(scrape_detail(detail_url)) + logging.info("get detail data %s", data) + save_data(data) + logging.info("save data successfully") + + +def mainByMulti(page): + index_html = scrape_index(page) + detail_urls = parse_index(index_html) + for detail_url in detail_urls: + data = parse_detail(scrape_detail(detail_url)) + logging.info("get detail data %s", data) + save_data(data) + logging.info("save data successfully") + + +if __name__ == '__main__': + # 单进程 + # main() + # 多进程 + import multiprocessing + + pool = multiprocessing.Pool() + pages = range(1, TOTAL_PAGE) + pool.map(mainByMulti, pages) + pool.close() + pool.join() diff --git a/Spider/爬虫基本库/__init__.py b/Spider/chapter02_爬虫基本库/__init__.py similarity index 100% rename from Spider/爬虫基本库/__init__.py rename to Spider/chapter02_爬虫基本库/__init__.py diff --git a/Spider/chapter02_爬虫基本库/httpx库_http2协议/__init__.py b/Spider/chapter02_爬虫基本库/httpx库_http2协议/__init__.py new file mode 100644 index 0000000..c6aeb28 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/httpx库_http2协议/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 18:59 +@Usage : +@Desc : +''' \ No newline at end of file diff --git a/Spider/chapter02_爬虫基本库/httpx库_http2协议/httpxLearning.py b/Spider/chapter02_爬虫基本库/httpx库_http2协议/httpxLearning.py new file mode 100644 index 0000000..8b50b09 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/httpx库_http2协议/httpxLearning.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 19:00 +@Usage : httpx库学习；request不支持http2.0相关协议。需要使用httpx。 +@Desc : httpx用法与request类似,具体内容参考:https://github.com/Python3WebSpider/HttpxTest +''' \ No newline at end of file diff --git a/Spider/chapter02_爬虫基本库/request库/__init__.py b/Spider/chapter02_爬虫基本库/request库/__init__.py new file mode 100644 index 0000000..47f4959 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/request库/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 15:51 +@Usage : +@Desc : +''' \ No newline at end of file diff --git a/Spider/chapter02_爬虫基本库/request库/requestLearning.py b/Spider/chapter02_爬虫基本库/request库/requestLearning.py new file mode 100644 index 0000000..23bb900 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/request库/requestLearning.py @@ -0,0 +1,219 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 15:51 +@Usage : +@Desc : request库学习 +''' + +import requests +import re +from requests.packages import urllib3 +from requests.auth import HTTPBasicAuth +from requests_oauthlib import OAuth1 + +''' +基本get使用 +''' + + +def get(): + data = { + 'name': 'germey', + 'age': 25 + } + response = requests.get('https://httpbin.org/get', params=data) + + print(response.text) + print(type(response.json())) # dict + print(response.json()) + pass + + +''' +抓取网页:使用模式匹配，抓取标题 +''' + + +def getPattern(): + response = requests.get('https://ssr1.scrape.center/') + pattern = '(.*?)' + pattern = re.compile(pattern, re.S) + titles = re.findall(pattern, response.text) + + print(titles) + + pass + + +''' +抓取二进制数据:使用模式匹配，抓取标题 +''' + + +def getBinary(): + response = requests.get('https://scrape.center/favicon.ico') + print(response.text) + print(response.content) + + with open('favicon.ico', 'wb') as f: + f.write(response.content) + + pass + + +''' +基本response的相关参数 +''' + + +def getResponse(): + response = requests.get('https://ssr1.scrape.center/') + + print(type(response.status_code), response.status_code) + print(type(response.headers), response.headers) + print(type(response.cookies), response.cookies) + print(type(response.history), response.history) + + exit() if not response.status_code == requests.codes.ok else print('Request Success!') + + +''' +基本post使用 +''' + + +def post(): + data = { + 'name': 'germey', + 'age': 25 + } + response = requests.post('https://httpbin.org/get', data=data) + + print(response.text) + + pass + + +''' +高级用法:上传文件 +''' + + +def postFile(): + file = { + 'file': open('favicon.ico', 'rb') + } + response = requests.post('https://httpbin.org/post', files=file) + + print(response.text) + + pass + + +''' +高级用法:cookie +cookie成功模拟了登录状态，这样就能爬取登录之后才能看到的页面了 +''' + + +def postCookie(): + response = requests.get('https://www.baidu.com') + print(response.cookies) + + for key, value in response.cookies.items(): + print(key, "=", value) + + pass + + +''' +Session维持: +如果第一次请求利用request库的post方法登录了某个网站，第二次想获取成功登录后自己的个人信息， +于是又用requests库的get方法区请求个人信息页面，这实际上相当于打开了两个浏览器，是两个完全独立的操作，这时需要维持Session +''' + + +def session(): + s = requests.Session() + s.get("https://www.httpbin.org/cookies/set/number/123456789") + r = s.get('https://www.httpbin.org/cookies') + print(r.text) # {"cookies": {"number": "123456789"}} + + +''' +SSL证书验证: +有些网站的HTTPS证书可能并不被CA机构认可，出现SSL证书错误 +''' + + +def SSL(): + # response = requests.get("https://ssr2.scrape.center/") + # print(response.status_code) # requests.exceptions.SSLError: HTTPSConnectionPool(host='ssr2.scrape.center', port=443): Max retries exceeded with url + + urllib3.disable_warnings() + response = requests.get("https://ssr2.scrape.center/", verify=False) + print(response.status_code) # 200 + + +''' +超时验证: +防止服务器不能即时响应 +''' + + +def timeout(): + # 设置超时时间1秒 + response = requests.get("https://www.httpbin.org/get", timeout=1) + # 如果不设置，则永久等待，如果设置为timeout=(5,30)则连接超时时间5秒，读取超时时间30秒 + print(response.status_code) + + +''' +身份认证: +在访问启用了基本身份认证的网站时，首先会弹出一个认证窗口 +''' + + +def Auth(): + # 用户名和密码都是admin + response = requests.get("https://ssr3.scrape.center/", auth=('admin', 'admin')) + + print(response.status_code) + + +''' +request还提供了其他认证方式，如OAuth认证，不过此时需要安装requests_oauthlib包 +''' +def OAuth(): + # 用户名和密码都是admin + url = 'https://api.twitter.com/1.1/account/verify_credentials.json' + auth = OAuth('your_app_key', 'your_app_sercet', 'user_oauth_token', 'user_oauth_token_secret') + response = requests.get(url, auth=auth) + + print(response.status_code) + + + +''' +代理设置: +某些网站请求几次可以正常获取内容。但一旦开始大规模爬取，可能弹出验证码或者挑战到登录认证页面等 +可以使用代理来解决这个问题 +''' +def proxy(): + # 用户名和密码都是admin + url = 'https://api.twitter.com/1.1/account/verify_credentials.json' + + proxies ={ + 'http':'http://10.10.10.10:1080', + 'https':'http://user:password@10.10.10.10:1080/' + } + + response = requests.get(url, proxy=proxies) + + print(response.status_code) + + +if __name__ == '__main__': + Auth() diff --git a/Spider/chapter02_爬虫基本库/request库/ssr1_scrape_center.html b/Spider/chapter02_爬虫基本库/request库/ssr1_scrape_center.html new file mode 100644 index 0000000..62df4f8 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/request库/ssr1_scrape_center.html @@ -0,0 +1,953 @@ + + + + + + + + Scrape | Movie + + + + + + + + +
+ + +
+
+
+ +
+
+
+
+ + + +
+
+ +

霸王别姬 - Farewell My Concubine

+ +
+ + + + + +
+
+ 中国内地、中国香港 + / + 171 分钟 +
+
+ + 1993-07-26 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

这个杀手不太冷 - Léon

+ +
+ + + + + + + +
+
+ 法国 + / + 110 分钟 +
+
+ + 1994-09-14 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

肖申克的救赎 - The Shawshank Redemption

+ +
+ + + + + +
+
+ 美国 + / + 142 分钟 +
+
+ + 1994-09-10 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

泰坦尼克号 - Titanic

+ +
+ + + + + + + +
+
+ 美国 + / + 194 分钟 +
+
+ + 1998-04-03 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

罗马假日 - Roman Holiday

+ +
+ + + + + + + +
+
+ 美国 + / + 118 分钟 +
+
+ + 1953-08-20 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

唐伯虎点秋香 - Flirting Scholar

+ +
+ + + + + + + +
+
+ 中国香港 + / + 102 分钟 +
+
+ + 1993-07-01 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

乱世佳人 - Gone with the Wind

+ +
+ + + + + + + + + +
+
+ 美国 + / + 238 分钟 +
+
+ + 1939-12-15 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

喜剧之王 - The King of Comedy

+ +
+ + + + + + + +
+
+ 中国香港 + / + 85 分钟 +
+
+ + 1999-02-13 上映 + +
+
+
+

+ 9.5

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

楚门的世界 - The Truman Show

+ +
+ + + + + +
+
+ 美国 + / + 103 分钟 +
+
+ +
+
+
+

+ 9.0

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+
+
+
+ + + +
+
+ +

狮子王 - The Lion King

+ +
+ + + + + + + +
+
+ 美国 + / + 89 分钟 +
+
+ + 1995-07-15 上映 + +
+
+
+

+ 9.0

+

+

+ + + + + + + + + + + + + + + +
+

+
+
+
+ +
+ +
+
+
+ +
+
+
+ +
+ \ No newline at end of file diff --git a/Spider/chapter02_爬虫基本库/re库_正则匹配/__init__.py b/Spider/chapter02_爬虫基本库/re库_正则匹配/__init__.py new file mode 100644 index 0000000..42223ba --- /dev/null +++ b/Spider/chapter02_爬虫基本库/re库_正则匹配/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 17:30 +@Usage : +@Desc : +''' \ No newline at end of file diff --git a/Spider/chapter02_爬虫基本库/re库_正则匹配/reLearning.py b/Spider/chapter02_爬虫基本库/re库_正则匹配/reLearning.py new file mode 100644 index 0000000..524dd54 --- /dev/null +++ b/Spider/chapter02_爬虫基本库/re库_正则匹配/reLearning.py @@ -0,0 +1,179 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/7 17:31 +@Usage : +@Desc : 正则匹配re库基本使用 +''' + +import re + +''' +基本使用 +''' + + +def baseUse(): + content = 'hello 123456789 World_This is a Reges Demo' + pattern = '^hello\s(\d+)\sWorld' + result = re.match(pattern, content) + + print(result) # + print(result.group()) # hello 123456789 World 输出匹配到的内容 + print(result.group(1)) # 123456789 输出第一个被()包围的匹配结果 + print(result.span()) # (0, 21) 输出匹配的范围 + + +# 高级用法 +''' +贪婪匹配与非贪婪匹配 +.*表示尽可能多匹配字符 +.*?表示尽可能少匹配字符 +''' + +''' +re.I 表示匹配对大小写不明感 +re.L 实现本地化识别匹配 +re.M 表示多航匹配，影响^和$ +re.S 表示匹配内容包括换行符在内的所有字符 +re.U 表示根据Unicode解析字符，这个表示会影响\w,\W,\b和\B +re.S 表示匹配内容包括换行符在内的所有字符 +''' +def prior(): + content = '''hello 123456789 World_This + is a Reges Demo + ''' + result = re.match('^he.*?(\d+).*?Demo$', content) + print(result.group(1)) # 未匹配到，报错 AttributeError: 'NoneType' object has no attribute 'group' + result = re.match('^he.*?(\d+).*?Demo$', content, re.S) + print(result.group(1)) # 123456789 + + +''' +search:模糊匹配 +''' +def search(): + content = 'Extra stings Hello 1234567 World_This is a Regex Demo Extra stings' + result = re.match('Hello.*?(\d+).*?Demo', content) # 必须要以Hello开头才能匹配到 + print(result) # None + result = re.search('Hello.*?(\d+).*?Demo', content) + print(result) # + + +def searchHtml(): + html = '''
+

经典老歌

+

+ 经典老歌列表 +

+ +
''' + result = re.search('(.*?)', html, re.S) + if result: + print(result.group(1), result.group(2)) + + result = re.search('(.*?)', html, re.S) + if result: + print(result.group(1), result.group(2)) + + result = re.search('(.*?)', html) + if result: + print(result.group(1), result.group(2)) + +''' +findAll:找到所有匹配的 +''' +def findall(): + html = '''
+

经典老歌

+

+ 经典老歌列表 +

+ +
''' + results = re.findall('(.*?)', html, re.S) + print(results) + print(type(results)) + for result in results: + print(result) + print(result[0], result[1], result[2]) + + results = re.findall('\s*?()?(\w+)()?\s*?', html, re.S) + for result in results: + print(result[1]) + + +''' +sub:正则匹配修改文本 +使用正则匹配，去除掉能匹配上的内容 +''' +def sub(): + html = '''
+

经典老歌

+

+ 经典老歌列表 +

+ +
''' + html = re.sub('|', '', html) + print(html) + results = re.findall('(.*?)', html, re.S) + for result in results: + print(result.strip()) + +''' +compile:编译 +将相关模式编译成pattern对象，进行复用 +''' +def complie(): + content1 = '2019-12-15 12:00' + content2 = '2019-12-17 12:55' + content3 = '2019-12-22 13:21' + pattern = re.compile('\d{2}:\d{2}') + result1 = re.sub(pattern, '', content1) + result2 = re.sub(pattern, '', content2) + result3 = re.sub(pattern, '', content3) + print(result1, result2, result3) + +if __name__ == '__main__': + sub() diff --git a/Spider/爬虫基本库/urllib库/__init__.py b/Spider/chapter02_爬虫基本库/urllib库/__init__.py similarity index 100% rename from Spider/爬虫基本库/urllib库/__init__.py rename to Spider/chapter02_爬虫基本库/urllib库/__init__.py diff --git a/Spider/爬虫基本库/urllib库/urllibLearning.py b/Spider/chapter02_爬虫基本库/urllib库/urllibLearning.py similarity index 100% rename from Spider/爬虫基本库/urllib库/urllibLearning.py rename to Spider/chapter02_爬虫基本库/urllib库/urllibLearning.py diff --git a/TensorFlow_eaxmple/Model_train_test/RUL/test/LSTMTest.py b/TensorFlow_eaxmple/Model_train_test/RUL/test/LSTMTest.py index a315dda..ee45e8b 100644 --- a/TensorFlow_eaxmple/Model_train_test/RUL/test/LSTMTest.py +++ b/TensorFlow_eaxmple/Model_train_test/RUL/test/LSTMTest.py @@ -27,7 +27,7 @@ batch_size = 32 EPOCH = 1000 unit = 512 # LSTM的维度 predict_num = 50 # 预测个数 -model_name = "LSTM" +model_name = "cnn_LSTM" save_name = r"self_{0}_hidden{1}_unit{2}_feature{3}_predict{4}.h5".format(model_name, hidden_num, unit, feature, predict_num) @@ -125,7 +125,8 @@ def predict_model(filter_num, dims): # LSTM = tf.keras.layers.LSTM(units=256, return_sequences=False)(LSTM) #### 自己 - LSTM = LSTMLayer(units=512, return_sequences=True)(input) + LSTM = tf.keras.layers.Conv1D(512, kernel_size=8, padding='same')(input) + LSTM = LSTMLayer(units=512, return_sequences=True)(LSTM) LSTM = LSTMLayer(units=256, return_sequences=False)(LSTM) x = tf.keras.layers.Dense(128, activation="relu")(LSTM) @@ -199,8 +200,7 @@ if __name__ == '__main__': # model.summary() # early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=100, mode='min', verbose=1) # - # history = model.fit(train_data, train_label_single, epochs=EPOCH, - # batch_size=batch_size, validation_data=(val_data, val_label_single), shuffle=True, verbose=2, + # history = model.fit(train_data, train_label_single, epochs=EPOCH, validation_data=(val_data, val_label_single), shuffle=True, verbose=1, # callbacks=[checkpoint, lr_scheduler, early_stop]) #### TODO 测试 diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/plot_loss.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/plot_loss.py new file mode 100644 index 0000000..44a5601 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/plot_loss.py @@ -0,0 +1,132 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/11/6 13:12 +@Usage : +@Desc : 画loss +''' + +import pandas as pd +import numpy as np +import matplotlib.pyplot as plt +from matplotlib.pyplot import rcParams + +import lttb + +file_name = "E:\论文写作\ISA论文返修\loss.txt" + + +def aux_loss(file_name="E:\论文写作\ISA论文返修\loss.txt"): + mse1_list = [] + mse2_list = [] + mse3_list = [] + total_list = [] + + with open(file_name, 'r', encoding='utf-8') as f: + i = 0 + for ann in f.readlines(): + ann = ann.strip('\n') # 去除文本中的换行符 + if i % 4 == 0: + # print(ann) + mse1 = float(ann.split(" ")[1]) + mse1_list.append(mse1) + elif i % 4 == 1: + mse2 = float(ann.split(" ")[1]) + mse2_list.append(mse2) + elif i % 4 == 2: + mse3 = float(ann.split(" ")[1]) + mse3_list.append(mse3) + elif i % 4 == 3: + total = float(ann.split(" ")[5]) + total_list.append(total) + i += 1 + + mse1_list = lttb.downsample(np.array([range(len(mse1_list)), mse1_list]).T, n_out=50) + mse2_list = lttb.downsample(np.array([range(len(mse2_list)), mse2_list]).T, n_out=50) + mse3_list = lttb.downsample(np.array([range(len(mse3_list)), mse3_list]).T, n_out=50) + total_list = lttb.downsample(np.array([range(len(total_list)), total_list]).T, n_out=50) + + plot_aux_loss(mse1_list, mse2_list, mse3_list) + + +def plot_aux_loss(mse1_list, mse2_list, mse3_list): + config = { + "font.family": 'Times New Roman', # 设置字体类型 + "axes.unicode_minus": False, # 解决负号无法显示的问题 + "axes.labelsize": 13 + } + rcParams.update(config) + pic1 = plt.figure(figsize=(8, 3), dpi=200) + # plt.ylim(0, 0.6) # 设置y轴范围 + plt.plot(mse1_list[:, 1], label='Predict Head 1 Loss') + plt.plot(mse2_list[:, 1], label='Predict Head 2 Loss') + plt.plot(mse3_list[:, 1], label='Predict Head 3 Loss') + plt.title('Training loss') + plt.xlabel('epoch') + plt.ylabel('loss') + plt.legend(loc='upper right') + plt.grid(alpha=0.8) + + plt.show() + + +def main_loss(file_name="E:\论文写作\ISA论文返修\main_loss.txt"): + mse1_list = [] + cross_Entropy_list = [] + total_list = [] + + with open(file_name, 'r', encoding='utf-8') as f: + i = 0 + for ann in f.readlines(): + ann = ann.strip('\n') # 去除文本中的换行符 + if i % 3 == 0: + # print(ann) + mse1 = float(ann.split(" ")[1]) + mse1_list.append(mse1) + elif i % 3 == 1: + cross_Entropy = float(ann.split(" ")[1])*0.02 + cross_Entropy_list.append(cross_Entropy) + elif i % 3 == 2: + total = float(ann.split(" ")[5]) + total_list.append(total) + i += 1 + + mse1_list = lttb.downsample(np.array([range(len(mse1_list)), mse1_list]).T, n_out=40) + cross_Entropy_list = lttb.downsample(np.array([range(len(cross_Entropy_list)), cross_Entropy_list]).T, n_out=40) + total_list = lttb.downsample(np.array([range(len(total_list)), total_list]).T, n_out=40) + + plot_main_loss(mse1_list, cross_Entropy_list,total_list) + +# 0.014995 0.016387 +# 0.014384 0.015866 +# 0.014985 0.014261 +# 0.013008 0.01349 +# 0.013285 0.013139 +# 0.012999 0.012714 +# 0.011451 0.012477 +# 0.010055 0.012471 +# 0.010303 0.014595 +def plot_main_loss(mse1_list, mse2_list,total_loss): + config = { + "font.family": 'Times New Roman', # 设置字体类型 + "axes.unicode_minus": False, # 解决负号无法显示的问题 + "axes.labelsize": 13 + } + rcParams.update(config) + pic1 = plt.figure(figsize=(8, 3), dpi=200) + # plt.ylim(0, 0.6) # 设置y轴范围 + plt.plot(mse1_list[:, 1], label='${L_{diff}}$') + plt.plot(mse2_list[:, 1], label='${L_{correct}}$') + plt.plot(total_loss[:, 1], label='${L_{main}}$') + plt.title('Training loss') + plt.xlabel('epoch') + plt.ylabel('loss') + plt.legend(loc='upper right') + plt.grid(alpha=0.8) + + plt.show() + + +if __name__ == '__main__': + main_loss() \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/test_plot.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/test_plot.py index 91b0933..fc652d1 100644 --- a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/test_plot.py +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/plot/test_plot.py @@ -56,12 +56,12 @@ def plot_result_banda(result_data): fig, ax = plt.subplots(1, 1) plt.rc('font', family='Times New Roman') # 全局字体样式 font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 5} # 设置坐标标签的字体大小，字体 - font2 = {'family': 'Times New Roman', 'weight': 'normal','size':7} # 设置坐标标签的字体大小，字体 + font2 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 7} # 设置坐标标签的字体大小，字体 plt.scatter(list(range(result_data.shape[0])), result_data, c='black', s=0.01, label="predict") # 画出 y=1 这条水平线 plt.axhline(0.5, c='red', label='Failure threshold', lw=1) # 箭头指向上面的水平线 - plt.axvline(result_data.shape[0] * 2 / 3-50, c='blue', ls='-.', lw=0.5, label='real fault') + plt.axvline(result_data.shape[0] * 2 / 3 - 50, c='blue', ls='-.', lw=0.5, label='real fault') # plt.axvline(415548, c='blue', ls='-.', lw=0.5, label='real fault') # 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.text(result_data.shape[0] * 5 / 6, 0.4, "Fault", fontsize=5, color='black', verticalalignment='top', @@ -85,8 +85,8 @@ def plot_result_banda(result_data): # pad调整label与坐标轴之间的距离 plt.tick_params(bottom=True, top=False, left=True, right=False, direction='inout', length=2, width=0.5, pad=1) # plt.yticks([index for index in indices1], classes1) - plt.ylabel('Confidence',fontdict=font2) - plt.xlabel('Time',fontdict=font2) + plt.ylabel('Confidence', fontdict=font2) + plt.xlabel('Time', fontdict=font2) plt.tight_layout() # plt.legend(loc='best', edgecolor='black', fontsize=4) plt.legend(loc='upper right', frameon=False, fontsize=4.5) @@ -98,11 +98,11 @@ def plot_result_banda(result_data): bbox_to_anchor=(0.1, 0.1, 1, 1), bbox_transform=ax.transAxes) axins.scatter(list(range(result_data.shape[0])), result_data, c='black', s=0.005, label="predict") - axins.axvline(result_data.shape[0] * 2 / 3-50, c='blue', ls='-.', lw=0.5, label='real fault') + axins.axvline(result_data.shape[0] * 2 / 3 - 50, c='blue', ls='-.', lw=0.5, label='real fault') plt.axhline(0.5, c='red', label='Failure threshold', lw=0.5) # 设置放大区间 # 设置放大区间 - zone_left = int(result_data.shape[0] * 2 / 3 -160) + zone_left = int(result_data.shape[0] * 2 / 3 - 160) zone_right = int(result_data.shape[0] * 2 / 3) + 40 # zone_left = int(result_data.shape[0] * 2 / 3 +250) # zone_right = int(result_data.shape[0] * 2 / 3) + 450 @@ -132,6 +132,7 @@ def plot_result_banda(result_data): plt.show() pass + def plot_result(result_data): parameters = { 'figure.dpi': 600, @@ -153,8 +154,7 @@ def plot_result(result_data): # 画出 y=1 这条水平线 plt.axhline(0.5, c='red', label='Failure threshold', lw=1) - - plt.axvline(result_data.shape[0] * 2 / 3-15, c='blue', ls='-.', lw=0.5, label='real fault') + plt.axvline(result_data.shape[0] * 2 / 3 - 15, c='blue', ls='-.', lw=0.5, label='real fault') plt.text(result_data.shape[0] * 5 / 6, 0.4, "Fault", fontsize=5, color='black', verticalalignment='top', horizontalalignment='center', bbox={'facecolor': 'grey', @@ -188,11 +188,11 @@ def plot_result(result_data): bbox_to_anchor=(0.1, 0.1, 1, 1), bbox_transform=ax.transAxes) axins.scatter(list(range(result_data.shape[0])), result_data, c='black', s=0.005, label="predict") - axins.axvline(result_data.shape[0] * 2 / 3-15, c='blue', ls='-.', lw=0.5, label='real fault') + axins.axvline(result_data.shape[0] * 2 / 3 - 15, c='blue', ls='-.', lw=0.5, label='real fault') plt.axhline(0.5, c='red', label='Failure threshold', lw=0.5) # 设置放大区间 # 设置放大区间 - zone_left = int(result_data.shape[0] * 2 / 3 -100) + zone_left = int(result_data.shape[0] * 2 / 3 - 100) zone_right = int(result_data.shape[0] * 2 / 3) + 100 x = list(range(result_data.shape[0])) @@ -221,6 +221,25 @@ def plot_result(result_data): pass +K = 18 +namuda = 0.01 + + +def EWMA(data, K=K, namuda=namuda): + # t是啥暂时未知 + length, = data.shape + t = 0 + data = pd.DataFrame(data).ewm(alpha=namuda).mean() + 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 data, np.broadcast_to(mid, shape=[length, ]), np.broadcast_to(UCL, shape=[length, ]), np.broadcast_to(LCL, + shape=[ + length, ]) + pass + + def plot_MSE(total_MSE, total_max): parameters = { 'figure.dpi': 600, @@ -251,12 +270,13 @@ def plot_MSE(total_MSE, total_max): classes = ['01/09/17', '08/09/17', '15/09/17', '22/09/17', '29/09/17'] plt.xticks([index + 0.5 for index in indices], classes, rotation=25) # 设置横坐标方向，rotation=45为45度倾斜 - plt.ylabel('Mse', fontsize=5) + plt.ylabel('MSE', fontsize=5) plt.xlabel('Time', fontsize=5) plt.tight_layout() - plt.plot(total_max, "--", label="max", linewidth=0.5) - plt.plot(total_MSE, label="mse", linewidth=0.5, color='purple') + plt.plot(total_max, "--", label="Threshold", linewidth=0.5, color='red') + plt.plot(total_max, "--", label="Threshold", linewidth=0.5, color='red') + plt.plot(total_MSE, label="MSE", linewidth=0.5, color='purple') plt.legend(loc='best', frameon=False, fontsize=5) # plt.plot(total_mean) @@ -265,6 +285,54 @@ def plot_MSE(total_MSE, total_max): pass +def plot_EWMA(model_name, total_MSE, total_mid, total_max, total_min): + parameters = { + 'figure.dpi': 600, + 'figure.figsize': (2.7, 2), + 'savefig.dpi': 600, + 'xtick.direction': 'in', + 'ytick.direction': 'in', + 'xtick.labelsize': 5, + 'ytick.labelsize': 5, + 'legend.fontsize': 5, + } + plt.rcParams.update(parameters) + plt.figure() + plt.rc('font', family='Times New Roman') # 全局字体样式#画混淆矩阵 + font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 5} # 设置坐标标签的字体大小，字体 + + result_data = total_MSE + # 画出 y=1 这条水平线 + # plt.axhline(0.5, c='red', label='Failure threshold',lw=1) + # 箭头指向上面的水平线 + # plt.arrow(result_data.shape[0]*2/3, 0.55, 2000, 0.085, width=0.00001, ec='red',length_includes_head=True) + # plt.text(result_data.shape[0] * 2 / 3 + 1000, 0.7, "real fault", fontsize=5, color='red', + # verticalalignment='top') + + plt.axvline(result_data.shape[0] * 2 / 3, c='blue', ls='-.', lw=0.5, label="real fault") + + indices = [result_data.shape[0] * i / 4 for i in range(5)] + classes = ['09/01', '09/08', '09/15', '09/22', '09/29/'] + + plt.xticks([index + 0.5 for index in indices], classes, rotation=25) # 设置横坐标方向，rotation=45为45度倾斜 + plt.ylabel('MSE', fontsize=5) + plt.xlabel('Time', fontsize=5) + plt.tight_layout() + + plt.plot(total_min, "--", label="LCL", linewidth=0.5, color='red') + plt.plot(total_mid, "--", label="mean(X)", linewidth=0.5) + plt.plot(total_max, "--", label="UCL", linewidth=0.5, color='red') + plt.plot(total_MSE, label="MSE", linewidth=0.5, color='purple') + plt.legend(loc='best', frameon=False, fontsize=5) + + # plt.plot(total_mean) + # plt.plot(min) + + plt.savefig('E:\论文写作\ISA论文返修\图片\比较方法的EWMD/{0}.png'.format(model_name)) + plt.show() + pass + + def plot_Corr(data, size: int = 1): parameters = { 'figure.dpi': 600, @@ -333,16 +401,21 @@ def plot_bar(y_data): plt.bar(x_width[1], y_data[1], lw=1, color=['#F5E3C4'], width=0.5, label="GRU", edgecolor='black') plt.bar(x_width[2], y_data[2], lw=1, color=['#EBC99D'], width=0.5, label="CNN-GRU", edgecolor='black') - plt.bar(x_width[3], y_data[3], lw=1, color=['#FFC79C'], width=0.5, label="DCConv", edgecolor='black') - plt.bar(x_width[4], y_data[4], lw=1, color=['#BEE9C7'], width=0.5, label="RepDCConv", edgecolor='black') - plt.bar(x_width[5], y_data[5], lw=1, color=['#B8E9D0'], width=0.5, label="RNet-MSE", edgecolor='black') - plt.bar(x_width[6], y_data[6], lw=1, color=['#B9E9E2'], width=0.5, label="RNet", edgecolor='black') - plt.bar(x_width[7], y_data[7], lw=1, color=['#D6E6F2'], width=0.5, label="RNet-SE", edgecolor='black') - plt.bar(x_width[8], y_data[8], lw=1, color=['#B4D1E9'], width=0.5, label="RNet-L", edgecolor='black') - plt.bar(x_width[9], y_data[9], lw=1, color=['#AEB5EE'], width=0.5, label="RNet-D", edgecolor='black') - plt.bar(x_width[10], y_data[10], lw=1, color=['#D2D3FC'], width=0.5, label="ResNet-18", edgecolor='black') - plt.bar(x_width[11], y_data[11], lw=1, color=['#D5A9FF'], width=0.5, label="ResNet-C", edgecolor='black') - plt.bar(x_width[12], y_data[12], lw=1, color=['#E000F5'], width=0.5, label="JMNet", edgecolor='black') + + plt.bar(x_width[3], y_data[13], lw=1, color=['#FCD58B'], width=0.5, label="CG-QA", edgecolor='black') + plt.bar(x_width[4], y_data[14], lw=1, color=['#FFBA75'], width=0.5, label="CG-3{0}".format(chr(963)), edgecolor='black') + + plt.bar(x_width[5], y_data[3], lw=1, color=['#FFC79C'], width=0.5, label="DCConv", edgecolor='black') + plt.bar(x_width[6], y_data[4], lw=1, color=['#BEE9C7'], width=0.5, label="RepDCConv", edgecolor='black') + plt.bar(x_width[7], y_data[5], lw=1, color=['#B8E9D0'], width=0.5, label="RNet-MSE", edgecolor='black') + plt.bar(x_width[8], y_data[6], lw=1, color=['#B9E9E2'], width=0.5, label="RNet", edgecolor='black') + plt.bar(x_width[9], y_data[7], lw=1, color=['#D6E6F2'], width=0.5, label="RNet-SE", edgecolor='black') + plt.bar(x_width[10], y_data[8], lw=1, color=['#B4D1E9'], width=0.5, label="RNet-L", edgecolor='black') + plt.bar(x_width[11], y_data[9], lw=1, color=['#AEB5EE'], width=0.5, label="RNet-D", edgecolor='black') + plt.bar(x_width[12], y_data[10], lw=1, color=['#D2D3FC'], width=0.5, label="ResNet-18", edgecolor='black') + plt.bar(x_width[13], y_data[11], lw=1, color=['#D5A9FF'], width=0.5, label="ResNet-C", edgecolor='black') + plt.bar(x_width[14], y_data[12], lw=1, color=['#E000F5'], width=0.5, label="JRFPN", edgecolor='black') + # plt.tick_params(bottom=False, top=False, left=True, right=False, direction='in', pad=1) plt.xticks([]) @@ -350,14 +423,14 @@ def plot_bar(y_data): plt.xlabel('Methods', fontsize=22) # plt.tight_layout() - num1, num2, num3, num4 = 0, 1, 3, 0 + num1, num2, num3, num4 = 0.015, 1, 3, 0 plt.legend(bbox_to_anchor=(num1, num2), loc=num3, borderaxespad=num4, ncol=5, frameon=False, handlelength=1, handletextpad=0.45, columnspacing=1) plt.ylim([-0.01, 5]) plt.show() -def acc(y_data=list): +def acc(y_data: list): parameters = { 'figure.dpi': 600, 'figure.figsize': (10, 6), @@ -373,22 +446,31 @@ def acc(y_data=list): plt.rc('font', family='Times New Roman') # 全局字体样式#画混淆矩阵 font1 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 5} # 设置坐标标签的字体大小，字体 - x_width = [i / 2 for i in range(0, len(y_data))] + x_width = [i for i in range(0, len(y_data))] # x2_width = [i + 0.3 for i in x_width] - plt.bar(x_width[0], y_data[0], lw=1, color=['#FAF4E1'], width=0.25, label="CNN", edgecolor='black') - plt.bar(x_width[1], y_data[1], lw=1, color=['#F5E3C4'], width=0.25, label="GRU", edgecolor='black') - plt.bar(x_width[2], y_data[2], lw=1, color=['#EBC99D'], width=0.25, label="CNN-GRU", edgecolor='black') - plt.bar(x_width[3], y_data[3], lw=1, color=['#FFC79C'], width=0.25, label="DCConv", edgecolor='black') - plt.bar(x_width[4], y_data[4], lw=1, color=['#BEE9C7'], width=0.25, label="RepDCConv", edgecolor='black') - plt.bar(x_width[5], y_data[5], lw=1, color=['#B8E9D0'], width=0.25, label="RNet-MSE", edgecolor='black') - plt.bar(x_width[6], y_data[6], lw=1, color=['#B9E9E2'], width=0.25, label="RNet", edgecolor='black') - plt.bar(x_width[7], y_data[7], lw=1, color=['#D6E6F2'], width=0.25, label="RNet-SE", edgecolor='black') - plt.bar(x_width[8], y_data[8], lw=1, color=['#B4D1E9'], width=0.25, label="RNet-L", edgecolor='black') - plt.bar(x_width[9], y_data[9], lw=1, color=['#AEB5EE'], width=0.25, label="RNet-D", edgecolor='black') - plt.bar(x_width[10], y_data[10], lw=1, color=['#D2D3FC'], width=0.25, label="ResNet-18", edgecolor='black') - plt.bar(x_width[11], y_data[11], lw=1, color=['#D5A9FF'], width=0.25, label="ResNet-C", edgecolor='black') - plt.bar(x_width[12], y_data[12], lw=1, color=['#E000F5'], width=0.25, label="JMNet", edgecolor='black') + _width = [i for i in range(0, len(y_data))] + # x2_width = [i + 0.3 for i in x_width] + + plt.bar(x_width[0], y_data[0], lw=1, color=['#FAF4E1'], width=0.5, label="CNN", edgecolor='black') + + plt.bar(x_width[1], y_data[1], lw=1, color=['#F5E3C4'], width=0.5, label="GRU", edgecolor='black') + plt.bar(x_width[2], y_data[2], lw=1, color=['#EBC99D'], width=0.5, label="CNN-GRU", edgecolor='black') + + plt.bar(x_width[3], y_data[13], lw=1, color=['#FCD58B'], width=0.5, label="CG-QA", edgecolor='black') + plt.bar(x_width[4], y_data[14], lw=1, color=['#FFBA75'], width=0.5, label="CG-3{0}".format(chr(963)), + edgecolor='black') + + plt.bar(x_width[5], y_data[3], lw=1, color=['#FFC79C'], width=0.5, label="DCConv", edgecolor='black') + plt.bar(x_width[6], y_data[4], lw=1, color=['#BEE9C7'], width=0.5, label="RepDCConv", edgecolor='black') + plt.bar(x_width[7], y_data[5], lw=1, color=['#B8E9D0'], width=0.5, label="RNet-MSE", edgecolor='black') + plt.bar(x_width[8], y_data[6], lw=1, color=['#B9E9E2'], width=0.5, label="RNet", edgecolor='black') + plt.bar(x_width[9], y_data[7], lw=1, color=['#D6E6F2'], width=0.5, label="RNet-SE", edgecolor='black') + plt.bar(x_width[10], y_data[8], lw=1, color=['#B4D1E9'], width=0.5, label="RNet-L", edgecolor='black') + plt.bar(x_width[11], y_data[9], lw=1, color=['#AEB5EE'], width=0.5, label="RNet-D", edgecolor='black') + plt.bar(x_width[12], y_data[10], lw=1, color=['#D2D3FC'], width=0.5, label="ResNet-18", edgecolor='black') + plt.bar(x_width[13], y_data[11], lw=1, color=['#D5A9FF'], width=0.5, label="ResNet-C", edgecolor='black') + plt.bar(x_width[14], y_data[12], lw=1, color=['#E000F5'], width=0.5, label="JRFPN", edgecolor='black') # plt.tick_params(bottom=False, top=False, left=True, right=False, direction='in', pad=1) plt.xticks([]) @@ -397,7 +479,8 @@ def acc(y_data=list): # plt.tight_layout() num1, num2, num3, num4 = 0, 1, 3, 0 - plt.legend(bbox_to_anchor=(num1, num2), loc=num3, borderaxespad=num4, ncol=5, frameon=False,handlelength=1,handletextpad=0.45,columnspacing=1) + plt.legend(bbox_to_anchor=(num1, num2), loc=num3, borderaxespad=num4, ncol=5, frameon=False, handlelength=1, + handletextpad=0.45, columnspacing=1) plt.ylim([60, 105]) plt.show() @@ -424,13 +507,18 @@ def plot_FNR1(y_data): plt.bar(x_width[0], y_data[0], lw=1, color=['#FAF4E1'], width=0.5 * 5 / 6, label="CNN", edgecolor='black') plt.bar(x_width[1], y_data[1], lw=1, color=['#F5E3C4'], width=0.5 * 5 / 6, label="GRU", edgecolor='black') plt.bar(x_width[2], y_data[2], lw=1, color=['#EBC99D'], width=0.5 * 5 / 6, label="CNN-GRU", edgecolor='black') - plt.bar(x_width[3], y_data[3], lw=1, color=['#FFC79C'], width=0.5 * 5 / 6, label="DCConv", edgecolor='black') - plt.bar(x_width[4], y_data[4], lw=1, color=['#BEE9C7'], width=0.5 * 5 / 6, label="RepDCConv", edgecolor='black') - plt.bar(x_width[5], y_data[5], lw=1, color=['#B8E9D0'], width=0.5 * 5 / 6, label="RNet-MSE", edgecolor='black') - plt.bar(x_width[6], y_data[6], lw=1, color=['#B9E9E2'], width=0.5 * 5 / 6, label="RNet", edgecolor='black') - plt.bar(x_width[7], y_data[7], lw=1, color=['#D6E6F2'], width=0.5 * 5 / 6, label="RNet-SE", edgecolor='black') - plt.bar(x_width[8], y_data[8], lw=1, color=['#B4D1E9'], width=0.5 * 5 / 6, label="RNet-L", edgecolor='black') - plt.bar(x_width[9], y_data[9], lw=1, color=['#AEB5EE'], width=0.5 * 5 / 6, label="RNet-D", edgecolor='black') + + plt.bar(x_width[3], y_data[10], lw=1, color=['#FCD58B'], width=0.5 * 5 / 6, label="CG-QA", edgecolor='black') + plt.bar(x_width[4], y_data[11], lw=1, color=['#EBC99D'], width=0.5 * 5 / 6, label="CG-3{0}".format(chr(963)), edgecolor='black') + + plt.bar(x_width[5], y_data[3], lw=1, color=['#FFC79C'], width=0.5 * 5 / 6, label="DCConv", edgecolor='black') + plt.bar(x_width[6], y_data[4], lw=1, color=['#BEE9C7'], width=0.5 * 5 / 6, label="RepDCConv", edgecolor='black') + plt.bar(x_width[7], y_data[5], lw=1, color=['#B8E9D0'], width=0.5 * 5 / 6, label="RNet-MSE", edgecolor='black') + plt.bar(x_width[8], y_data[6], lw=1, color=['#B9E9E2'], width=0.5 * 5 / 6, label="RNet", edgecolor='black') + plt.bar(x_width[9], y_data[7], lw=1, color=['#D6E6F2'], width=0.5 * 5 / 6, label="RNet-SE", edgecolor='black') + plt.bar(x_width[10], y_data[8], lw=1, color=['#B4D1E9'], width=0.5 * 5 / 6, label="RNet-L", edgecolor='black') + plt.bar(x_width[11], y_data[9], lw=1, color=['#AEB5EE'], width=0.5 * 5 / 6, label="RNet-D", edgecolor='black') + # plt.tick_params(bottom=False, top=False, left=True, right=False, direction='in', pad=1) plt.xticks([]) @@ -440,7 +528,7 @@ def plot_FNR1(y_data): plt.xlabel('Methods', fontsize=22) # plt.tight_layout() - num1, num2, num3, num4 = 0.05, 1, 3, 0 + num1, num2, num3, num4 = 0.025, 1, 3, 0 plt.legend(bbox_to_anchor=(num1, num2), loc=num3, borderaxespad=num4, ncol=5, frameon=False, handlelength=1, handletextpad=0.45, columnspacing=1) @@ -479,15 +567,25 @@ def plot_FNR2(y_data): # plt.bar(x_width[8], y_data[8], lw=1, color=['#E000F5'], width=0.5 * 2 / 3, label="JMNet", edgecolor='black') plt.bar(x_width[0], y_data[0], lw=1, color=['#AEB5EE'], width=0.5 * 2 / 3, label="ResNet-18", edgecolor='black') - plt.bar(x_width[1], y_data[1], color=['#FFFFFF'], label=" ") - plt.bar(x_width[2], y_data[2], color=['#FFFFFF'], label=" ") - plt.bar(x_width[3], y_data[3], color=['#FFFFFF'], label=" ") - plt.bar(x_width[4], y_data[4],lw=1, color=['#D5A9FF'], width=0.5 * 2 / 3, label="RNet-C", edgecolor='black') - plt.bar(x_width[5], y_data[5], color=['#FFFFFF'], label=" ") - plt.bar(x_width[6], y_data[6], color=['#FFFFFF'], label=" ") + plt.bar(x_width[1], y_data[1], color=['#FFFFFF'], label=" ") + plt.bar(x_width[2], y_data[2], color=['#FFFFFF'], label=" ") + plt.bar(x_width[3], y_data[3], color=['#FFFFFF'], label=" ") + plt.bar(x_width[4], y_data[4], color=['#FFFFFF'], label=" ") + plt.bar(x_width[5], y_data[5], color=['#FFFFFF'], label=" ") + + + plt.bar(x_width[6], y_data[6], lw=1, color=['#D5A9FF'], width=0.5 * 2 / 3, label="RNet-C", edgecolor='black') + plt.bar(x_width[7], y_data[7], color=['#FFFFFF'], label=" ") + plt.bar(x_width[8], y_data[8], color=['#FFFFFF'], label=" ") + plt.bar(x_width[9], y_data[9], color=['#FFFFFF'], label=" ") + plt.bar(x_width[10], y_data[10], color=['#FFFFFF'], label=" ") + plt.bar(x_width[11], y_data[11], color=['#FFFFFF'], label=" ") + # plt.bar(x_width[7] + 2.0, y_data[10], lw=0.5, color=['#8085e9'], width=1, label="ResNet-18", edgecolor='black') - plt.bar(x_width[7], y_data[7], color=['#FFFFFF'], label=" ") - plt.bar(x_width[8], y_data[8],lw=1, color=['#E000F5'], width=0.5 * 2 / 3, label="JMNet", edgecolor='black') + + + plt.bar(x_width[12], y_data[12], lw=1, color=['#E000F5'], width=0.5 * 2 / 3, label="JRFPN", edgecolor='black') + plt.bar(x_width[13], y_data[13], color=['#FFFFFF'], label=" ") # plt.tick_params(bottom=False, top=False, left=True, right=False, direction='in', pad=1) plt.xticks([]) @@ -554,14 +652,14 @@ def plot_hot_one(data): pass -def plot_mse(file_name_mse="../others_idea/mse",data:str=''): +def plot_mse(file_name_mse="../others_idea/mse", data: str = ''): mse = np.loadtxt(file_name_mse, delimiter=",") - raw_data=np.loadtxt(data,delimiter=",") - raw_data=raw_data[:,:mse.shape[1]] - print("mse:",mse.shape) - print("raw_data:",raw_data.shape) - res=raw_data-mse - mse.shape[0]*2/3 + raw_data = np.loadtxt(data, delimiter=",") + raw_data = raw_data[:, :mse.shape[1]] + print("mse:", mse.shape) + print("raw_data:", raw_data.shape) + res = raw_data - mse + mse.shape[0] * 2 / 3 # mse = mse[2000:2300] # mse = mse[1800:2150] @@ -589,20 +687,19 @@ def plot_mse(file_name_mse="../others_idea/mse",data:str=''): plt.axvline(res.shape[1] * 2 / 3, c='purple', ls='-.', lw=0.5, label="real fault") plt.plot(res[i, :], lw=0.5) - - plt.show() + def plot_mse_single(file_name_mse="../self_try/compare/mse/JM_banda/banda_joint_result_predict1.csv"): mse = np.loadtxt(file_name_mse, delimiter=",") - print("mse:",mse.shape) + print("mse:", mse.shape) - need_shape=int(mse.shape[0]*2/3) + need_shape = int(mse.shape[0] * 2 / 3) # mse = mse[2000:2300] # mse = mse[1800:2150] # mse = mse[ need_shape+100:need_shape+377] - mse = mse[ need_shape-300:need_shape-10] + mse = mse[need_shape - 300:need_shape - 10] parameters = { 'figure.dpi': 600, @@ -613,7 +710,7 @@ def plot_mse_single(file_name_mse="../self_try/compare/mse/JM_banda/banda_joint_ 'xtick.labelsize': 6, 'ytick.labelsize': 6, 'legend.fontsize': 5, - 'font.family':'Times New Roman' + 'font.family': 'Times New Roman' } plt.rcParams.update(parameters) font2 = {'family': 'Times New Roman', 'weight': 'normal', 'size': 7} # 设置坐标标签的字体大小，字体 @@ -622,9 +719,9 @@ def plot_mse_single(file_name_mse="../self_try/compare/mse/JM_banda/banda_joint_ plt.rc('font', family='Times New Roman') # 全局字体样式#画混淆矩阵 indices = [mse.shape[0] * i / 4 for i in range(5)] # classes = ['07:21','08:21', '09:21', '10:21', '11:21'] - classes = ['01:58','02:58', '03:58', '04:58', '05:58'] + classes = ['01:58', '02:58', '03:58', '04:58', '05:58'] - plt.xticks([index for index in indices], classes, rotation=25) # 设置横坐标方向，rotation=45为45度倾斜 + plt.xticks([index for index in indices], classes, rotation=25) # 设置横坐标方向，rotation=45为45度倾斜 # pad调整label与坐标轴之间的距离 plt.tick_params(bottom=True, top=False, left=True, right=False, direction='inout', length=2, width=0.5, pad=1) plt.ylabel('Residual', fontdict=font2) @@ -632,8 +729,6 @@ def plot_mse_single(file_name_mse="../self_try/compare/mse/JM_banda/banda_joint_ plt.tight_layout() plt.plot(mse[:], lw=0.5) - - plt.show() @@ -641,37 +736,36 @@ def plot_3d(): # 线 fig = plt.figure() ax = fig.add_subplot(projection='3d') - data,label = read_data(file_name='G:\data\SCADA数据\jb4q_8_delete_total_zero.csv',isNew=False) + data, label = read_data(file_name='G:\data\SCADA数据\jb4q_8_delete_total_zero.csv', isNew=False) print(data) print(data.shape) - x=range(data.shape[1]) - y=range(data.shape[0]) - color=[[0.16,0.14,0.13], - [0.89,0.09,0.05], - [0.12,0.56,1.00], - [0.01,0.66,0.62], - [0.63,0.04,0.83], - [0.63,0.32,0.18], - [0.20,0.63,0.79], - [0.50,0.16,0.16], - [0.61,0.4,0.12], - [1.00,0.38,0.00], - [0.53,0.81,0.92], - [0.13,0.55,0.13], - [1.00,0.89,0.52], - [0.44,0.50,0.41], - [0.20,0.63,0.79], - [0.00,0.78,0.55], - [1.00,0.39,0.28], - [0.25,0.41,0.88]] + x = range(data.shape[1]) + y = range(data.shape[0]) + color = [[0.16, 0.14, 0.13], + [0.89, 0.09, 0.05], + [0.12, 0.56, 1.00], + [0.01, 0.66, 0.62], + [0.63, 0.04, 0.83], + [0.63, 0.32, 0.18], + [0.20, 0.63, 0.79], + [0.50, 0.16, 0.16], + [0.61, 0.4, 0.12], + [1.00, 0.38, 0.00], + [0.53, 0.81, 0.92], + [0.13, 0.55, 0.13], + [1.00, 0.89, 0.52], + [0.44, 0.50, 0.41], + [0.20, 0.63, 0.79], + [0.00, 0.78, 0.55], + [1.00, 0.39, 0.28], + [0.25, 0.41, 0.88]] # c颜色，marker：样式*雪花 - ax.plot( xs=y,ys=x, zs=data, c=color, marker="*") + ax.plot(xs=y, ys=x, zs=data, c=color, marker="*") plt.show() - def test_result(file_name: str = result_file_name): # result_data = np.recfromcsv(file_name) result_data = np.loadtxt(file_name, delimiter=",") @@ -680,7 +774,7 @@ def test_result(file_name: str = result_file_name): theshold = len(result_data) print(theshold) print(theshold * 2 / 3) - theshold = theshold * 2 / 3-50 + theshold = theshold * 2 / 3 - 50 # 计算误报率和漏报率 positive_rate = result_data[:int(theshold)][result_data[:int(theshold)] < 0.66].__len__() / ( theshold * 2 / 3) @@ -783,22 +877,166 @@ def test_model_visualization(model_name=file_name): plot_hot(needed_data) +def plot_loss(): + from matplotlib import rcParams + + config = { + "font.family": 'Times New Roman', # 设置字体类型 + "axes.unicode_minus": False, # 解决负号无法显示的问题 + "axes.labelsize": 13 + } + rcParams.update(config) + + pic1 = plt.figure(figsize=(8, 6), dpi=200) + plt.subplot(211) + plt.plot(np.arange(1, epochs + 1), src_acc_list, 'b', label='TrainAcc') + plt.plot(np.arange(1, epochs + 1), val_acc_list, 'r', label='ValAcc') + plt.ylim(0.3, 1.0) # 设置y轴范围 + plt.title('Training & Validation accuracy') + plt.xlabel('epoch') + plt.ylabel('accuracy') + plt.legend(loc='lower right') + plt.grid(alpha=0.4) + + plt.subplot(212) + plt.plot(np.arange(1, epochs + 1), train_loss_list, 'b', label='TrainLoss') + plt.plot(np.arange(1, epochs + 1), val_loss_list, 'r', label='ValLoss') + plt.ylim(0, 0.08) # 设置y轴范围 + plt.title('Training & Validation loss') + plt.xlabel('epoch') + plt.ylabel('loss') + plt.legend(loc='upper right') + plt.grid(alpha=0.4) + + # 获取当前时间戳 + timestamp = int(time.time()) + + # 将时间戳转换为字符串 + timestamp_str = str(timestamp) + plt.savefig(timestamp_str, dpi=200) + if __name__ == '__main__': + # model_list = ["DCConv", "GRU"] + # + # for model_name in model_list: + # # model_name = "CNN_GRU" + # mse = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}/banda\mse.csv'.format( + # model_name), + # delimiter=',') + # max = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}/banda\max.csv'.format( + # model_name), + # delimiter=',') + # data, mid, UCL, LCL = EWMA(mse) + # + # plot_EWMA( + # model_name=model_name+"_banda", + # total_MSE=data, + # total_mid=mid, + # total_max=UCL, + # total_min=LCL + # ) + # + # model_list = ["RNet_D", "RNet_L","RNet_S","RNet_MSE"] + # predict_list = ["predict1","predict2","predict3"] + # + # for model_name in model_list: + # for predict_head in predict_list: + # # model_name = "RNet_L" + # # predict_head = "predict3" + # mse = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}/banda\{0}_banda_mse_{1}.csv'.format( + # model_name, predict_head), + # delimiter=',') + # max = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}/banda\{0}_banda_max_{1}.csv'.format( + # model_name, predict_head), + # delimiter=',') + # data, mid, UCL, LCL = EWMA(mse) + # + # plot_EWMA( + # model_name=model_name + predict_head+"_banda", + # total_MSE=data, + # total_mid=mid, + # total_max=UCL, + # total_min=LCL + # ) + + # model_name = "RNet_D" + # predict_head = "predict1" + # mse = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}\{0}_timestamp120_feature10_mse_{1}.csv'.format( + # model_name, predict_head), + # delimiter=',') + # max = np.loadtxt( + # 'E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\{0}\{0}_timestamp120_feature10_max_{1}.csv'.format( + # model_name, predict_head), + # delimiter=',') + # data, mid, UCL, LCL = EWMA(mse) + # + # plot_EWMA( + # model_name=model_name + predict_head , + # total_MSE=data, + # total_mid=mid, + # total_max=UCL, + # total_min=LCL + # ) + + # test_result("E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\RNet_345\RNet_345_timestamp120_feature10_result.csv") # test_mse(fi) # test_result( # file_name='E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\ResNet\ResNet_timestamp120_feature10_result.csv') # test_corr() # acc() - # list = [3.77, 2.64, 2.35, 2.05, 1.76, 1.09, 0.757, 0.82, 1.1, 0.58, 0, 0.03, 0.02] - # test_bar(list) - - # list=[98.56,98.95,99.95,96.1,95,99.65,76.25,72.64,75.87,68.74] - # plot_FNR1(list) - # # - # list=[3.43,1.99,1.92,2.17,1.63,1.81,1.78,1.8,0.6] - list=[3.43,1.99,1.92,2.17,1.8,1.81,1.78,1.8,0.6] - plot_FNR2(list) + # list = [ + # 3.77, + # 2.64, + # 2.35, + # + # 2.05, + # 1.76, + # 1.09, + # 0.76, + # 0.82, + # 1.10, + # 0.58, + # 0, + # 0.03, + # 0.02, + # 0.21, + # 2.18, + # ] + # + # # test_bar(list) + # list = [ + # 64.63, + # 65.26, + # 65.11, + # + # 66.6, + # 67.15, + # 73.86, + # 66.28, + # 75.24, + # 73.98, + # 76.7, + # 98.86, + # 99.38, + # 99.79, + # 70.41, + # 66.71 + # ] + # # acc(list) + # + list=[98.56,98.95,99.95,96.1,95,99.65,76.25,72.64,75.87,68.74,88.36,95.52] + plot_FNR1(list) + # # # # + # # list=[3.43,1.99,1.92,2.17,1.63,1.81,1.78,1.8,0.6] + # list=[3.43,0,0,0,0,0,1.8,0,0,0,0,0,0.6,0] + # # list = [98.56, 98.95, 99.95, 96.1, 95, 99.65, 76.25, 72.64, 75.87, 68.74, 88.36, 95.52] + # plot_FNR2(list) # 查看网络某一层的权重 # test_model_visualization(model_name = "E:\跑模型\论文写作/SE.txt") @@ -810,11 +1048,11 @@ if __name__ == '__main__': # 单独预测图 # plot_mse('E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\JM_banda/banda_joint_result_predict3.csv',data='E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\JM_banda/raw_data.csv') # plot_mse_single('E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\RNet_D/banda\RNet_D_banda_mse_predict1.csv') - #画3d图 + # plot_mse_single('E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\RNet_3\RNet_3_timestamp120_feature10_result.csv') + # 画3d图 # plot_3d() - - #原始数据图 + # 原始数据图 # file_names='E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse\JM_banda/raw_data.csv' # data= np.loadtxt(file_names,delimiter=',') diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/Joint_Monitoring_banda.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/Joint_Monitoring_banda.py index e4d0966..bbfd514 100644 --- a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/Joint_Monitoring_banda.py +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/Joint_Monitoring_banda.py @@ -7,7 +7,6 @@ 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_daban as loadData from model.Joint_Monitoring.Joint_Monitoring_banda import Joint_Monitoring @@ -40,7 +39,7 @@ save_name = "../hard_model/weight/{0}_epoch16_0.0009_0.0014/weight".format(model feature_num, batch_size, EPOCH) -save_step_two_name = "../hard_model/two_weight/{0}_epoch24_9875_9867/weight".format(model_name, +save_step_two_name = "../hard_model/two_weight/temp{0}/weight".format(model_name, time_stamp, feature_num, batch_size, @@ -385,7 +384,7 @@ def train_step_one(train_data, train_label1, train_label2): 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()) + SaveBestModel(model=model, save_name=save_name, history_loss=history_val_loss, loss_value=val_loss.numpy(),is_all=True) # 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()) @@ -586,28 +585,32 @@ if __name__ == '__main__': #### TODO 第一步训练 # 单次测试 # train_step_one(train_data=train_data_healthy[:128, :, :], train_label1=train_label1_healthy[:128, :],train_label2=train_label2_healthy[:128, ]) + # 整体训练 # 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) + + #### 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 - # 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)) + 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)) ### TODO 测试测试集 # step_one_model = Joint_Monitoring() 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 aa036de..a0f8efc 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 @@ -18,7 +18,7 @@ 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 model.ChannelAttention.Dynamic_channelAttention import DynamicChannelAttention from condition_monitoring.data_deal import loadData_daban as loadData from model.Joint_Monitoring.compare.RNet_C import Joint_Monitoring diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet.py index 93d7827..82e89da 100644 --- a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet.py +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/self_try/compare/RNet.py @@ -15,7 +15,8 @@ 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 +import time + from condition_monitoring.data_deal import loadData_daban as loadData from model.Joint_Monitoring.compare.RNet import Joint_Monitoring @@ -591,10 +592,21 @@ if __name__ == '__main__': # 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_one_model.build(input_shape=(30,120,10)) + # step_one_model.summary() # # step_two_model = Joint_Monitoring() # step_two_model.load_weights(save_name) + start = time.time() + # 中间写上代码块 + + step_one_model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + # #### TODO 计算MSE healthy_size, _, _ = train_data_healthy.shape unhealthy_size, _, _ = train_data_unhealthy.shape 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 index ca3e9f2..1f37482 100644 --- 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 @@ -8,7 +8,7 @@ import seaborn as sns from sklearn.model_selection import train_test_split from condition_monitoring.data_deal import loadData_daban as loadData from keras.callbacks import EarlyStopping - +import time '''超参数设置''' time_stamp = 120 feature_num = 10 @@ -68,22 +68,7 @@ unhealthy_patience = 5 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 @@ -328,6 +313,14 @@ if __name__ == '__main__': # model.save("./model/ResNet.h5") model = tf.keras.models.load_model("model/ResNet_banda/ResNet_banda_epoch10_9884.h5") + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + # 结果展示 healthy_size, _, _ = train_data_healthy.shape diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Cnn-gru.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Cnn-gru.py new file mode 100644 index 0000000..47435de --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Cnn-gru.py @@ -0,0 +1,496 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，与JointNet相同深度,进行预测 +@Desc : +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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/DCConv/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/DCConv/banda/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''' +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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 = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM = tf.keras.layers.Conv1D(10, 3)(input) + LSTM = tf.keras.layers.Conv1D(20, 3)(LSTM) + LSTM = tf.keras.layers.GRU(20, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(40, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(80, return_sequences=False)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=64)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=128)(LSTM) + # LSTM = tf.keras.layers.Conv1D(40, 3, padding="causal",dilation_rate=2)(LSTM) + + # LSTM = LSTM[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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(1,9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + + missNum = mse1[mse1[:] < max[0]].__len__() + + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + + + 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 + model = DCConv_Model() + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + + # history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + # train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + # batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + # callbacks=[checkpoint, lr_scheduler, early_stop]) + + ## TODO testing + # # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + # model = tf.keras.models.load_model(save_name) + # # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + # + # start = time.time() + # # 中间写上代码块 + # + # model.predict(train_data_healthy, batch_size=32) + # end = time.time() + # print("data_size:", train_data_healthy.shape) + # print('Running time: %s Seconds' % (end - start)) + # + 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) + + newModel = tf.keras.models.load_model(save_name) + # 单次测试 + # getResult(newModel, + # 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(newModel, 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=False) + # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Conv.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Conv.py new file mode 100644 index 0000000..4fa1d19 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/Conv.py @@ -0,0 +1,497 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，与JointNet相同深度,进行预测 +@Desc : +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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 = "E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse/CNN/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "E:\self_example\TensorFlow_eaxmple\Model_train_test\condition_monitoring\self_try\compare\mse/CNN/banda/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''' +'''文件名''' +'''文件名''' +file_name = "G:\data\SCADA数据\jb4q_8_delete_all_zero.csv" + +''' +文件说明：jb4q_8_delete_all_zero.csv是删除了除异常以外的所有0值的文件 +文件从0:300454行均是正常值(2019/7.30 00:00:00 - 2019/9/18 11:21:00) +从300455:317052行均是异常值(2019/9/18 11:21:01 - 2019/9/29 23:59:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 300454 +# 最后异常的时间点 +unhealthy_date = 317052 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM = tf.keras.layers.Conv1D(10, 3)(input) + LSTM = tf.keras.layers.Conv1D(20, 3)(LSTM) + LSTM = tf.keras.layers.Conv1D(20, 3)(LSTM) + LSTM = tf.keras.layers.Conv1D(40, 3)(LSTM) + LSTM = tf.keras.layers.Conv1D(80, 3)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=64)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=128)(LSTM) + # LSTM = tf.keras.layers.Conv1D(40, 3, padding="causal",dilation_rate=2)(LSTM) + + LSTM = LSTM[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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(1,9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + + missNum = mse1[mse1[:] < max[0]].__len__() + + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + + + 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 + model = DCConv_Model() + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + + history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + callbacks=[checkpoint, lr_scheduler, early_stop]) + + ## TODO testing + # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + # newModel = tf.keras.models.load_model(save_name) + # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + + 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) + + newModel = tf.keras.models.load_model(save_name) + # 单次测试 + # getResult(newModel, + # 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(newModel, 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) + # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/DCConv.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/DCConv.py new file mode 100644 index 0000000..58d1b9c --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/DCConv.py @@ -0,0 +1,496 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，与JointNet相同深度,进行预测 +@Desc : +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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/DCConv/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/DCConv/banda/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''' +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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 = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=2)(input) + LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=4)(LSTM) + LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=8)(LSTM) + LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=16)(LSTM) + LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=32)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=64)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=128)(LSTM) + # LSTM = tf.keras.layers.Conv1D(40, 3, padding="causal",dilation_rate=2)(LSTM) + + LSTM = LSTM[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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(1,9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + + missNum = mse1[mse1[:] < max[0]].__len__() + + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + + + 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 + model = DCConv_Model() + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + + history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + callbacks=[checkpoint, lr_scheduler, early_stop]) + + ## TODO testing + # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + # newModel = tf.keras.models.load_model(save_name) + # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + + 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) + + newModel = tf.keras.models.load_model(save_name) + # 单次测试 + # getResult(newModel, + # 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(newModel, 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) + # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/GRU.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/GRU.py new file mode 100644 index 0000000..256d277 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/GRU.py @@ -0,0 +1,493 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，与JointNet相同深度,进行预测 +@Desc : +''' + + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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/DCConv/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/DCConv/banda/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''' +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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 = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM = tf.keras.layers.GRU(10, return_sequences=True)(input) + LSTM = tf.keras.layers.GRU(20, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(20, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(40, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(80, return_sequences=False)(LSTM) + + # LSTM = LSTM[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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(1,9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + + missNum = mse1[mse1[:] < max[0]].__len__() + + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + + + 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 + model = DCConv_Model() + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + + # history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + # train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + # batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + # callbacks=[checkpoint, lr_scheduler, early_stop]) + + ## TODO testing + # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + model = tf.keras.models.load_model(save_name) + # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + + 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) + + newModel = tf.keras.models.load_model(save_name) + # 单次测试 + # getResult(newModel, + # 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(newModel, 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) + # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/RepDCConv.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/RepDCConv.py new file mode 100644 index 0000000..a437166 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/RepDCConv.py @@ -0,0 +1,520 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，与JointNet相同深度,进行预测 +@Desc : +''' + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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/DCConv/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/DCConv/banda/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''' +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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 = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM1 = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=2)(input) + LSTM2 = tf.keras.layers.Conv1D(10, 2, padding="causal", dilation_rate=2)(input) + LSTM3 = tf.keras.layers.Conv1D(10, 1, padding="causal", dilation_rate=2)(input) + + t1 = tf.add(tf.add(LSTM1, LSTM2), LSTM3) + + LSTM1 = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=4)(t1) + LSTM2 = tf.keras.layers.Conv1D(10, 2, padding="causal", dilation_rate=4)(t1) + LSTM3 = tf.keras.layers.Conv1D(10, 1, padding="causal", dilation_rate=4)(t1) + + t2 = tf.add(tf.add(LSTM1, LSTM2), LSTM3) + + + + LSTM1 = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=8)(t2) + LSTM2 = tf.keras.layers.Conv1D(10, 2, padding="causal", dilation_rate=8)(t2) + LSTM3 = tf.keras.layers.Conv1D(10, 1, padding="causal", dilation_rate=8)(t2) + + t3 = tf.add(tf.add(LSTM1, LSTM2), LSTM3) + + + LSTM1 = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=16)(t3) + LSTM2 = tf.keras.layers.Conv1D(10, 2, padding="causal", dilation_rate=16)(t3) + LSTM3 = tf.keras.layers.Conv1D(10, 1, padding="causal", dilation_rate=16)(t3) + + t4 = tf.add(tf.add(LSTM1, LSTM2), LSTM3) + + + + LSTM1 = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=32)(t4) + LSTM2 = tf.keras.layers.Conv1D(10, 2, padding="causal", dilation_rate=32)(t4) + LSTM3 = tf.keras.layers.Conv1D(10, 1, padding="causal", dilation_rate=32)(t4) + + t5 = tf.add(tf.add(LSTM1, LSTM2), LSTM3) + + + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=64)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=128)(LSTM) + # LSTM = tf.keras.layers.Conv1D(40, 3, padding="causal",dilation_rate=2)(LSTM) + + LSTM = t5[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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, 9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + missNum = mse1[mse1[:] < max[0]].__len__() + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + 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 + model = DCConv_Model() + + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + # early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + # + # history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + # train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + # batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + # callbacks=[checkpoint, lr_scheduler, early_stop]) + + ## TODO testing + # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + newModel = tf.keras.models.load_model(save_name) + # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + + 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) + + newModel = tf.keras.models.load_model(save_name) + # 单次测试 + # getResult(newModel, + # 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(newModel, 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) + # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/__init__.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/__init__.py new file mode 100644 index 0000000..f8b166b --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 10:43 +@Usage : +@Desc : +''' \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/Transformer.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/Transformer.py new file mode 100644 index 0000000..00417bc --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/Transformer.py @@ -0,0 +1,101 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 10:43 +@Usage : +@Desc : 计算transformer的参数和时间复杂度 6层self_attention +''' +import tensorflow as tf +from tensorflow.keras import Model, layers, initializers +import numpy as np +from model.SelfAttention.SelfAttention import Block + + +class Transformer(Model): + + # depth表示的是重复encoder block的次数，num_heads表示的是在multi-head self-attention中head的个数 + # MLP block中有一个Pre_logist,这里指的是，当在较大的数据集上学习的时候Pre_logist就表示一个全连接层加上一个tanh激活函数 + # 当在较小的数据集上学习的时候，Pre_logist是没有的，而这里的representation_size表示的就是Pre_logist中全连接层的节点个数 + # num_classes表示分类的类数 + def __init__(self, embed_dim=768, + depth=12, num_heads=12, qkv_bias=True, qk_scale=None, + drop_ratio=0., attn_drop_ratio=0., drop_path_ratio=0., + representation_size=None, num_classes=1000, name="ViT-B/16"): + super(Transformer, self).__init__(name=name) + + self.embed_dim = embed_dim + self.depth = depth + self.num_heads = num_heads + self.qkv_bias = qkv_bias + self.qk_scale = qk_scale + self.drop_ratio = drop_ratio + self.attn_drop_ratio = attn_drop_ratio + self.drop_path_ratio = drop_path_ratio + self.representation_size = representation_size + self.num_classes = num_classes + + dpr = np.linspace(0., drop_path_ratio, depth) # stochastic depth decay rule + # 用一个for循环重复Block模块 + # 在用droppath时的drop_path_ratio是由0慢慢递增到我们所指定的drop_path_ratio的 + # 所以我们在构建Block时，这里的drop_path_ratio时变化的，所以用 np.linspace方法创建一个等差数列来初始化drop_path_ratio + self.blocks = [Block(dim=embed_dim, num_heads=num_heads, qkv_bias=qkv_bias, + qk_scale=qk_scale, drop_ratio=drop_ratio, attn_drop_ratio=attn_drop_ratio, + drop_path_ratio=dpr[i], name="encoderblock_{}".format(i)) + for i in range(depth)] + + self.norm = layers.LayerNormalization(epsilon=1e-6, name="encoder_norm") + + # 接下来，如果传入了representation_size，就构建一个全连接层，激活函数为tanh + # 如果没有传入的话，就不做任何操作 + if representation_size: + self.has_logits = True + self.pre_logits = layers.Dense(representation_size, activation="tanh", name="pre_logits") + else: + self.has_logits = False + self.pre_logits = layers.Activation("linear") + + # 定义最后一个全连接层，节点个数就是我们的分类个数num_classes + self.head = layers.Dense(num_classes, name="head", kernel_initializer=initializers.Zeros()) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'embed_dim': self.embed_dim, + 'depth': self.depth, + 'num_heads': self.num_heads, + 'qkv_bias': self.qkv_bias, + 'qk_scale': self.qk_scale, + 'drop_ratio': self.drop_ratio, + 'attn_drop_ratio': self.attn_drop_ratio, + 'drop_path_ratio': self.drop_path_ratio, + 'representation_size': self.representation_size, + 'num_classes': self.num_classes + } + ) + base_config = super(Transformer, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, training=None): + # [B, H, W, C] -> [B, num_patches, embed_dim] + x = inputs # [B, 196, 768] + + for block in self.blocks: + x = block(x, training=training) + + x = self.norm(x) + # 这里是提取class_toke的输出，然后用切片的方式，而刚刚是将class_toke拼接在最前面的 + # 所以这里用切片的方式，去取class_toke的输出，并将它传递给pre_logits + x = self.pre_logits(x[:, 0]) + # 最后传递给head + x = self.head(x) + # 为什么只用class_toke对应的输出，而不用每一个patches对应的输出呢？ + # 可以参考原文bird 网络 + + return x + + +if __name__ == '__main__': + # 使用方式 + Transformer(embed_dim=10, depth=8, num_heads=1, num_classes=10) diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/__init__.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/__init__.py new file mode 100644 index 0000000..f8b166b --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/complete/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 10:43 +@Usage : +@Desc : +''' \ No newline at end of file diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/poor-Cnn-gru.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/poor-Cnn-gru.py new file mode 100644 index 0000000..33b9a13 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/poor-Cnn-gru.py @@ -0,0 +1,569 @@ +# -*- coding: utf-8 -*- + +# coding: utf-8 + + +''' +@Author : dingjiawen +@Date : 2022/10/11 18:52 +@Usage : 对比实验，使用四分位图技术 +@Desc : +''' + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt + +from condition_monitoring.data_deal import loadData_daban as loadData +from model.Joint_Monitoring.Joint_Monitoring3 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 +from keras.callbacks import EarlyStopping +import random +import time + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "DCConv" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +save_name = "./trianed/{0}_{1}_{2}.h5".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/DCConv/banda/mse.csv".format(model_name, + time_stamp, + feature_num, + batch_size, + EPOCH) +save_max_name = "./mse/DCConv/banda/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''' +'''文件名''' +save_mse_name=r"./compare/mse/JM_banda/{0}_result.csv".format(model_name) +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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0:96748行均是正常值(2019/12.30 00:00:00 - 2020/3/11 05:58:00) +从96748:107116行均是异常值(2020/3/11 05:58:01 - 2021/3/18 11:04:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 + + +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 DCConv_Model(): + input = tf.keras.Input(shape=[time_stamp, feature_num]) + input = tf.cast(input, tf.float32) + + LSTM = tf.keras.layers.Conv1D(10, 3)(input) + LSTM = tf.keras.layers.Conv1D(20, 3)(LSTM) + LSTM = tf.keras.layers.GRU(20, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(40, return_sequences=True)(LSTM) + LSTM = tf.keras.layers.GRU(80, return_sequences=False)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=64)(LSTM) + # LSTM = tf.keras.layers.Conv1D(10, 3, padding="causal", dilation_rate=128)(LSTM) + # LSTM = tf.keras.layers.Conv1D(40, 3, padding="causal",dilation_rate=2)(LSTM) + + # LSTM = LSTM[:, -1, :] + # bn = tf.keras.layers.BatchNormalization()(LSTM) + + # d1 = tf.keras.layers.Dense(20)(LSTM) + # bn = tf.keras.layers.BatchNormalization()(d1) + + output = tf.keras.layers.Dense(128, name='output1')(LSTM) + output = tf.keras.layers.Dense(10, name='output')(output) + model = tf.keras.Model(inputs=input, outputs=output) + return model + pass + + +def get_MSE(data, label, new_model, isStandard: bool = True, isPlot: bool = True, predictI: int = 1): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + if isStandard: + dims, = mse.shape + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + print("max:", max) + # 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, 9)) + plt.plot(max) + plt.plot(mse) + plt.plot(mean) + # plt.plot(min) + plt.show() + else: + if isPlot: + plt.figure(random.randint(1, 9)) + plt.plot(mse) + # plt.plot(min) + plt.show() + return mse + + return mse, mean, max + # 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 = True, predictI: int = 1): + # TODO 计算MSE确定阈值 + # TODO 计算MSE确定阈值 + + mse, mean, max = get_MSE(healthy_data, healthy_label, model) + + # 误报率的计算 + total, = mse.shape + faultNum = 0 + faultList = [] + faultNum = mse[mse[:] > max[0]].__len__() + # 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 + mse1 = get_MSE(unhealthy_data, unhealthy_label, model, isStandard=False) + + 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 + save_max_name1 = save_max_name + + np.savetxt(save_mse_name1, total_mse, delimiter=',') + np.savetxt(save_max_name1, total_max, delimiter=',') + + all, = mse1.shape + + missNum = mse1[mse1[:] < max[0]].__len__() + + print("all:", all) + miss_rate = missNum / all + print("漏报率:", miss_rate) + + plt.figure(random.randint(1, 100)) + plt.plot(total_max) + plt.plot(total_mse) + plt.plot(total_mean) + # plt.plot(min) + plt.show() + pass + + +def iqr_outliers(df: pd.DataFrame): + q1 = df.quantile(0.25) + q3 = df.quantile(0.75) + iqr = q3 - q1 + Lower_tail = q1 - 1.5 * iqr + Upper_tail = q3 + 1.5 * iqr + outlier = [] + for i in df.iloc[:, 0]: + if i > float(Upper_tail) or i < float(Lower_tail): # float限定 + outlier.append(i) + print("Outliers:", outlier) + + +def iqr_outliers_np_all(df): + length, feature = df.shape + toatl_Outliers = set() + for i in range(feature): + cur = df[:, i] + cur = np.array(cur) + q1 = np.percentile(cur, [25]) + q3 = np.percentile(cur, [75]) + iqr = q3 - q1 + Lower_tail = q1 - 1.5 * iqr + Upper_tail = q3 + 1.5 * iqr + cur_Outliers = [] + for i in range(len(cur)): + if cur[i] > float(Upper_tail) or cur[i] < float(Lower_tail): # float限定 + toatl_Outliers.add(i) + cur_Outliers.append(i) + print("cur_Outliers.shape:", len(cur_Outliers)) + print("cur_Outliers:", cur_Outliers) + print("Outliers.shape:", len(toatl_Outliers)) + print("Outliers:", toatl_Outliers) + unhealthy_outlier = [] + for s in toatl_Outliers: + if s >= healthy_date: + unhealthy_outlier.append(s) + print("unhealthy_outlier.shape:", len(unhealthy_outlier)) + print("unhealthy_outlier.shape:", len(unhealthy_outlier)/(unhealthy_date-healthy_date)) + print("unhealthy_outlier:", unhealthy_outlier) + return sorted(unhealthy_outlier) + + +def sigma_outliers_np_all(df): + length, feature = df.shape + toatl_Outliers = set() + for i in range(feature): + cur = df[:, i] + cur = np.array(cur) + mean = np.mean(cur[:200000,]) + sigma = np.sqrt(np.var(cur[:200000,])) + + Lower_tail = mean - 3 * sigma + Upper_tail = mean + 3 * sigma + cur_Outliers = [] + for i in range(len(cur)): + if cur[i] > float(Upper_tail) or cur[i] < float(Lower_tail): # float限定 + toatl_Outliers.add(i) + cur_Outliers.append(i) + print("cur_Outliers.shape:", len(cur_Outliers)) + print("cur_Outliers:", cur_Outliers) + print("Outliers.shape:", len(toatl_Outliers)) + print("Outliers:", toatl_Outliers) + unhealthy_outlier = [] + for s in toatl_Outliers: + if s >= healthy_date: + unhealthy_outlier.append(s) + print("unhealthy_outlier.shape:", len(unhealthy_outlier)) + print("unhealthy_outlier.shape:", len(unhealthy_outlier)/(unhealthy_date-healthy_date)) + print("unhealthy_outlier:", unhealthy_outlier) + return sorted(unhealthy_outlier) + + +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) + + total_train_data = total_data[:unhealthy_date, :] + # unhealthy_outlier = iqr_outliers_np_all(total_train_data) + unhealthy_outlier = sigma_outliers_np_all(total_train_data) + # #### TODO 第一步训练 + # # 单次测试 + # model = DCConv_Model() + # + # checkpoint = tf.keras.callbacks.ModelCheckpoint( + # filepath=save_name, + # monitor='val_loss', + # verbose=2, + # save_best_only=True, + # mode='min') + # lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + # + # model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + # model.build(input_shape=(batch_size, time_stamp, feature_num)) + # model.summary() + # early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + # + # # history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + # # train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + # # batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + # # callbacks=[checkpoint, lr_scheduler, early_stop]) + # + # ## TODO testing + # # # test_data, test_label = get_training_data(total_data[:healthy_date, :]) + # # model = tf.keras.models.load_model(save_name) + # # # mse, mean, max = get_MSE(test_data, test_label, new_model=newModel) + # # + # # start = time.time() + # # # 中间写上代码块 + # # + # # model.predict(train_data_healthy, batch_size=32) + # # end = time.time() + # # print("data_size:", train_data_healthy.shape) + # # print('Running time: %s Seconds' % (end - start)) + # # + # 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) + # + # newModel = tf.keras.models.load_model(save_name) + # # 单次测试 + # # getResult(newModel, + # # 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(newModel, 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=False) + # # mse, mean, max = get_MSE(train_data_healthy[healthy_size - 2 * unhealthy_size:, :], + # # train_label1_healthy[healthy_size - 2 * unhealthy_size:, :], new_model=newModel) + pass diff --git a/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/transformer_complete.py b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/transformer_complete.py new file mode 100644 index 0000000..0ae78e2 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/condition_monitoring/返修/transformer_complete.py @@ -0,0 +1,285 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 14:23 +@Usage : +@Desc : +''' + +import tensorflow as tf +import tensorflow.keras +import numpy as np +import pandas as pd +import matplotlib.pyplot as plt +from condition_monitoring.data_deal import loadData_daban as loadData +# from model.Joint_Monitoring.Joint_Monitoring_banda 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 +from condition_monitoring.返修.complete.Transformer import Transformer +from model.SelfAttention.SelfAttention import Block +from keras.callbacks import EarlyStopping +import time + +'''超参数设置''' +time_stamp = 120 +feature_num = 10 +batch_size = 32 +learning_rate = 0.001 +EPOCH = 101 +model_name = "transformer" +'''EWMA超参数''' +K = 18 +namuda = 0.01 +'''保存名称''' + +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) + +save_mse_name = r"./compare/mse/JM_banda/{0}_result.csv".format(model_name) +'''文件名''' +file_name = "G:\data\SCADA数据\SCADA_已处理_粤水电达坂城2020.1月-5月\风机15.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0:96748行均是正常值(2019/12.30 00:00:00 - 2020/3/11 05:58:00) +从96748:107116行均是异常值(2020/3/11 05:58:01 - 2021/3/18 11:04:00) +''' +'''文件参数''' +# 最后正常的时间点 +healthy_date = 96748 +# 最后异常的时间点 +unhealthy_date = 107116 +# 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 + + +# 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 get_MSE(data, label, new_model): + predicted_data = new_model.predict(data) + + temp = np.abs(predicted_data - label) + temp1 = (temp - np.broadcast_to(np.mean(temp, axis=0), shape=predicted_data.shape)) + temp2 = np.broadcast_to(np.sqrt(np.var(temp, axis=0)), shape=predicted_data.shape) + temp3 = temp1 / temp2 + mse = np.sum((temp1 / temp2) ** 2, axis=1) + print("z:", mse) + print(mse.shape) + + # mse=np.mean((predicted_data-label)**2,axis=1) + print("mse", mse) + + dims, = mse.shape + + mean = np.mean(mse) + std = np.sqrt(np.var(mse)) + max = mean + 3 * std + # 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, ]) + + # plt.plot(max) + # plt.plot(mse) + # plt.plot(mean) + # # plt.plot(min) + # plt.show() + # + # + return mse, mean, max + # 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 + + +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) + + +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 + + ####### TODO 训练 + model = Transformer(embed_dim=10, depth=5, num_heads=1, num_classes=10,representation_size=128) + checkpoint = tf.keras.callbacks.ModelCheckpoint( + filepath=save_name, + monitor='val_loss', + verbose=2, + save_best_only=True, + mode='min') + lr_scheduler = tf.keras.callbacks.ReduceLROnPlateau(monitor='val_loss', factor=0.2, patience=3, min_lr=0.001) + + model.compile(optimizer=tf.optimizers.Adam(), loss=tf.losses.mse) + model.build(input_shape=(batch_size, time_stamp, feature_num)) + model.summary() + early_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, mode='min', verbose=1) + + history = model.fit(train_data_healthy[:train_data_healthy.shape[0] // 7, :, :], + train_label1_healthy[:train_label1_healthy.shape[0] // 7, ], epochs=EPOCH, + batch_size=batch_size * 10, validation_split=0.2, shuffle=True, verbose=1, + callbacks=[checkpoint, lr_scheduler, early_stop]) + # + # + # + # #### TODO 测试 + + # + start = time.time() + # 中间写上代码块 + + model.predict(train_data_healthy, batch_size=32) + end = time.time() + print("data_size:", train_data_healthy.shape) + print('Running time: %s Seconds' % (end - start)) + + trained_model = tf.keras.models.load_model(save_name, custom_objects={'Block': Block}) + # + # + # + # # 使用已知的点进行预测 + # + # pass diff --git a/TensorFlow_eaxmple/Model_train_test/model/ChannelAttention/Dynamic_channelAttention.py b/TensorFlow_eaxmple/Model_train_test/model/ChannelAttention/Dynamic_channelAttention.py index 3cc7dba..db9054c 100644 --- a/TensorFlow_eaxmple/Model_train_test/model/ChannelAttention/Dynamic_channelAttention.py +++ b/TensorFlow_eaxmple/Model_train_test/model/ChannelAttention/Dynamic_channelAttention.py @@ -97,6 +97,16 @@ class DynamicPooling(layers.Layer): self.pool_size = pool_size pass + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'pool_size': self.pool_size + } + ) + base_config = super(DynamicPooling, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + def build(self, input_shape): if len(input_shape) != 3: raise ValueError('Inputs to `DynamicChannelAttention` should have rank 3. ' diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring3.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring3.py index 22d362a..cf83f17 100644 --- a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring3.py +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring3.py @@ -15,7 +15,6 @@ 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 import math diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring_banda.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring_banda.py index f3acba7..5133bb8 100644 --- a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring_banda.py +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/Joint_Monitoring_banda.py @@ -13,13 +13,13 @@ 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.ChannelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling from model.LossFunction.smooth_L1_Loss import SmoothL1Loss import math +MSE_loss1_list=[] +MSE_loss2_list=[] +MSE_loss3_list=[] class Joint_Monitoring(keras.Model): @@ -28,12 +28,13 @@ class Joint_Monitoring(keras.Model): 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',kernel_initializer=0.7,bias_initializer=1) + # self.conv1 = tf.keras.layers.Conv1D(filters=conv_filter, kernel_size=1, strides=2, padding='SAME',kernel_initializer=0.7,bias_initializer=1) + 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',kernel_initializer=0.7,bias_initializer=1) + 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() @@ -255,6 +256,11 @@ class Joint_Monitoring(keras.Model): print("MSE_loss1:", MSE_loss1.numpy()) print("MSE_loss2:", MSE_loss2.numpy()) print("MSE_loss3:", MSE_loss3.numpy()) + + # MSE_loss1_list.append(MSE_loss1.numpy()) + # MSE_loss2_list.append(MSE_loss2.numpy()) + # MSE_loss3_list.append(MSE_loss3.numpy()) + loss = MSE_loss1 + MSE_loss2 + MSE_loss3 Accuracy_num = 0 diff --git a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet.py b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet.py index 96626c6..d71190f 100644 --- a/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet.py +++ b/TensorFlow_eaxmple/Model_train_test/model/Joint_Monitoring/compare/RNet.py @@ -15,7 +15,7 @@ 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 model.ChannelAttention.Dynamic_channelAttention import DynamicChannelAttention, DynamicPooling from condition_monitoring.data_deal import loadData from model.LossFunction.smooth_L1_Loss import SmoothL1Loss diff --git a/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/SelfAttention.py b/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/SelfAttention.py new file mode 100644 index 0000000..d1a7896 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/SelfAttention.py @@ -0,0 +1,417 @@ +# -*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 10:53 +@Usage : 即插即用的self_attention模块 +@Desc : +''' +import tensorflow as tf +from tensorflow.keras import Model, layers, initializers +import numpy as np + + +class PatchEmbed(layers.Layer): + """ + 2D Image to Patch Embedding + """ + + def __init__(self, img_size=224, patch_size=16, embed_dim=768): + super(PatchEmbed, self).__init__() + self.embed_dim = embed_dim + self.patch_size=patch_size + self.img_size = (img_size, img_size) + # 将图片划分成 img_size // patch_size行img_size // patch_size列的网格 + self.grid_size = (img_size // patch_size, img_size // patch_size) + # 行列相乘，可以得到patch的数目，即14*14=196，到时候flatten层的输入即是196维 + self.num_patches = self.grid_size[0] * self.grid_size[1] + # con2d是16*16，然后步距也是16，即可将图片划分 + self.proj = layers.Conv2D(filters=embed_dim, kernel_size=patch_size, + strides=patch_size, padding='SAME', + kernel_initializer=initializers.LecunNormal(), + bias_initializer=initializers.Zeros()) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'img_size': self.img_size[0], + 'patch_size': self.patch_size, + 'embed_dim': self.embed_dim + } + ) + base_config = super(PatchEmbed, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + # 正向传播过程 + def call(self, inputs, **kwargs): + # B-Batch的维度，H-高度, W-宽度, C-channel的维度 + B, H, W, C = inputs.shape + # 设置的高和宽要和上面的保持一致，如果不一致会报错 + # 这里和CNN模型不一样，CNN模型会有一个全局池化层来对图片大小进行一个池化，所以可以不限制图片的具体大小 + # 而在transformer中是需要加上一个position embedding的，而这个层就需要知道图片的具体维度来设置position embedding的大小 + assert H == self.img_size[0] and W == self.img_size[1], \ + f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})." + # 将input传入设置好的conv2d，对其进行输出 + x = self.proj(inputs) + # [B, H, W, C] -> [B, H*W, C] + # reshape成可以进入flatten的形式 + x = tf.reshape(x, [B, self.num_patches, self.embed_dim]) + return x + + +class ConcatClassTokenAddPosEmbed(layers.Layer): + def __init__(self, embed_dim=768, num_patches=196, name=None): + super(ConcatClassTokenAddPosEmbed, self).__init__(name=name) + self.embed_dim = embed_dim + self.num_patches = num_patches + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'num_patches': self.num_patches, + 'embed_dim': self.embed_dim + } + ) + base_config = super(ConcatClassTokenAddPosEmbed, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def build(self, input_shape): + # 创建两个可训练的参数weight,对应于cls_token，pos_embed的weight + # shape第一维是batch维度，后面分别是1*768和197*768，trainable表示是可训练的参数 + self.cls_token = self.add_weight(name="cls", + shape=[1, 1, self.embed_dim], + initializer=initializers.Zeros(), + trainable=True, + dtype=tf.float32) + self.pos_embed = self.add_weight(name="pos_embed", + shape=[1, self.num_patches + 1, self.embed_dim], + initializer=initializers.RandomNormal(stddev=0.02), + trainable=True, + dtype=tf.float32) + + def call(self, inputs, **kwargs): + # 需要获取一下batch的，因为输入图片的时候不是一张图片，而是将图片打包成一个个batch同时输入进去 + batch_size, _, _ = inputs.shape + + # [1, 1, 768] -> [B, 1, 768] + # 把cls_token复制B份进行拼接，broadcast_to方法，将cls_token进行一下广播，在batch维度就可以变成batch维度 + cls_token = tf.broadcast_to(self.cls_token, shape=[batch_size, 1, self.embed_dim]) + # 与input进行拼接 + x = tf.concat([cls_token, inputs], axis=1) # [B, 197, 768] + # 加上位置编码 + x = x + self.pos_embed + + return x + + +class SelfAttention(layers.Layer): + # 定义两个权重初始化方法，方便后续调用 + k_ini = initializers.GlorotUniform() + b_ini = initializers.Zeros() + + # dim是前面的embed的dimension,num_heads多头注意力的头数 + def __init__(self, + dim, + num_heads=8, + qkv_bias=False, + qk_scale=None, + attn_drop_ratio=0., + proj_drop_ratio=0., + name=None): + super(SelfAttention, self).__init__(name=name) + + self.dim = dim + self.num_heads = num_heads + self.qkv_bias = qkv_bias + self.qk_scale = qk_scale + self.attn_drop_ratio = attn_drop_ratio + self.proj_drop_ratio = proj_drop_ratio + + + + + # 每一个head的dimension=输入的dimension/num_heads + head_dim = dim // num_heads + # 做softmax时会除一个sqrt(dk),这里的scale就是那个sqrt(dk)——缩放因子 + # 如果传入了qk_scale就用传入的，如果没传入就用sqrt(head_dim) + self.scale = qk_scale or head_dim ** -0.5 + # 在有的做法中，生成QKV三个矩阵时会用三个全连接层 + # 这里用3*dim维的大全连接层，可以达到一样的效果 + self.qkv = layers.Dense(dim * 3, use_bias=qkv_bias, name="qkv", + kernel_initializer=self.k_ini, bias_initializer=self.b_ini) + self.attn_drop = layers.Dropout(attn_drop_ratio) + # 这里的全连接层是生成Wo矩阵，将得到的b进一步拼接 + # 由于这里multi-head self-attention模块的输入输出维度是一样的，所以这里的节点个数是dim + self.proj = layers.Dense(dim, name="out", + kernel_initializer=self.k_ini, bias_initializer=self.b_ini) + self.proj_drop = layers.Dropout(proj_drop_ratio) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'dim': self.dim, + 'num_heads': self.num_heads, + 'qkv_bias': self.qkv_bias, + 'qk_scale': self.qk_scale, + 'attn_drop_ratio': self.attn_drop_ratio, + 'proj_drop_ratio': self.proj_drop_ratio, + } + ) + base_config = super(SelfAttention, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, training=None): + # 由于进入self-attention的时候就已经将展平过了，所以这里的inputs.shape只有三个维度 + # [batch_size, num_patches + 1, total_embed_dim] + B, N, C = inputs.shape + # B, N, C = tf.shape(inputs)[0],tf.shape(inputs)[1],tf.shape(inputs)[2] + + # qkv(): -> [batch_size, num_patches + 1, 3 * total_embed_dim] + qkv = self.qkv(inputs) + # 分成三份，分别对应qkv,C // self.num_heads得到每一个head的维度 + # reshape: -> [batch_size, num_patches + 1, 3, num_heads, embed_dim_per_head] + # qkv = tf.reshape(qkv, shape=[B, N, 3, self.num_heads, C // self.num_heads]) + qkv = tf.reshape(qkv, shape=[-1, N, 3, self.num_heads, C // self.num_heads]) + # qkv = tf.keras.layers.Reshape(target_shape=[B, N, 3, self.num_heads, C // self.num_heads])(qkv) + # 用transpose方法，来调整一下维度的顺序，[2, 0, 3, 1, 4]表示调换之后的顺序 + # transpose: -> [3, batch_size, num_heads, num_patches + 1, embed_dim_per_head] + qkv = tf.transpose(qkv, [2, 0, 3, 1, 4]) + # [batch_size, num_heads, num_patches + 1, embed_dim_per_head] + q, k, v = qkv[0], qkv[1], qkv[2] + + # transpose: -> [batch_size, num_heads, embed_dim_per_head, num_patches + 1] + # 这里的矩阵相乘实际上指的是矩阵的最后两个维度相乘，而b的转置(transpose_b) + # 实际上是[batch_size, num_heads, embed_dim_per_head, num_patches + 1] + # multiply -> [batch_size, num_heads, num_patches + 1, num_patches + 1] + attn = tf.matmul(a=q, b=k, transpose_b=True) * self.scale + attn = tf.nn.softmax(attn, axis=-1) + attn = self.attn_drop(attn, training=training) + + # 与v相乘得到b + # multiply -> [batch_size, num_heads, num_patches + 1, embed_dim_per_head] + x = tf.matmul(attn, v) + # 再用transpose调换一下顺序 + # transpose: -> [batch_size, num_patches + 1, num_heads, embed_dim_per_head] + x = tf.transpose(x, [0, 2, 1, 3]) + # reshape: -> [batch_size, num_patches + 1, total_embed_dim] + # x = tf.reshape(x, [B, N, C]) + x = tf.reshape(x, [-1, N, C]) + + # 与Wo相乘，进一步融合得到输出 + x = self.proj(x) + x = self.proj_drop(x, training=training) + return x + + +class MLP(layers.Layer): + """ + MLP as used in Vision Transformer, MLP-Mixer and related networks + """ + # 定义两个权重初始化方法 + k_ini = initializers.GlorotUniform() + b_ini = initializers.RandomNormal(stddev=1e-6) + + # in_deatures表示输入MLP模块对应的dimension，mlp_ratio=4.0表示翻四倍 + def __init__(self, in_features, mlp_ratio=4.0, drop_rate=0., name=None): + super(MLP, self).__init__(name=name) + + self.in_features = in_features + self.mlp_ratio = mlp_ratio + self.drop_rate = drop_rate + + self.fc1 = layers.Dense(int(in_features * mlp_ratio), name="Dense_0", + kernel_initializer=self.k_ini, bias_initializer=self.b_ini) + self.act = layers.Activation("relu") + self.fc2 = layers.Dense(in_features, name="Dense_1", + kernel_initializer=self.k_ini, bias_initializer=self.b_ini) + self.drop = layers.Dropout(drop_rate) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'in_features': self.in_features, + 'mlp_ratio': self.mlp_ratio, + 'drop_rate': self.drop_rate, + } + ) + base_config = super(MLP, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, training=None): + x = self.fc1(inputs) + x = self.act(x) + x = self.drop(x, training=training) + x = self.fc2(x) + x = self.drop(x, training=training) + return x + + +class Block(layers.Layer): + def __init__(self, + dim, + num_heads=8, + qkv_bias=False, + qk_scale=None, + drop_ratio=0., + attn_drop_ratio=0., + drop_path_ratio=0., + name=None): + super(Block, self).__init__(name=name) + + self.dim = dim + self.num_heads = num_heads + self.qkv_bias = qkv_bias + self.qk_scale = qk_scale + self.drop_ratio = drop_ratio + self.attn_drop_ratio = attn_drop_ratio + self.drop_path_ratio = drop_path_ratio + + + + + # LayerNormalization来进行正则化 + self.norm1 = layers.LayerNormalization(epsilon=1e-6, name="LayerNorm_0") + # 调用Attention类，来实现MultiHeadAttention + self.attn = SelfAttention(dim, num_heads=num_heads, + qkv_bias=qkv_bias, qk_scale=qk_scale, + attn_drop_ratio=attn_drop_ratio, proj_drop_ratio=drop_ratio, + name="MultiHeadAttention") + # 这里所一个判断，如果drop_path_ratio > 0.就构建droppath方法 + # 如果drop_path_ratio < 0,就用linear,即输入是什么输出就是什么，不作操作 + # droppath方式的实现Dropout+noise_shape=(None, 1, 1)就可以实现droppath方法 + # 第一个None表示的是batch维度，第一个1表示的是num_patches+1,最后一个1表示的是embed_dimension + # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here + self.drop_path = layers.Dropout(rate=drop_path_ratio, noise_shape=(None, 1, 1)) if drop_path_ratio > 0. \ + else layers.Activation("linear") + self.norm2 = layers.LayerNormalization(epsilon=1e-6, name="LayerNorm_1") + self.mlp = MLP(dim, drop_rate=drop_ratio, name="MlpBlock") + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'dim': self.dim, + 'num_heads': self.num_heads, + 'qkv_bias': self.qkv_bias, + 'qk_scale': self.qk_scale, + 'drop_ratio': self.drop_ratio, + 'attn_drop_ratio': self.attn_drop_ratio, + 'drop_path_ratio': self.drop_path_ratio, + } + ) + base_config = super(Block, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + def call(self, inputs, training=None): + # 对应于图中第一个加号以前的部分 + x = inputs + self.drop_path(self.attn(self.norm1(inputs)), training=training) + # 对应于图中第一个加号以后的部分 + x = x + self.drop_path(self.mlp(self.norm2(x)), training=training) + return x + + +class VisionTransformer(Model): + + # depth表示的是重复encoder block的次数，num_heads表示的是在multi-head self-attention中head的个数 + # MLP block中有一个Pre_logist,这里指的是，当在较大的数据集上学习的时候Pre_logist就表示一个全连接层加上一个tanh激活函数 + # 当在较小的数据集上学习的时候，Pre_logist是没有的，而这里的representation_size表示的就是Pre_logist中全连接层的节点个数 + # num_classes表示分类的类数 + def __init__(self, img_size=224, patch_size=16, embed_dim=768, + depth=12, num_heads=12, qkv_bias=True, qk_scale=None, + drop_ratio=0., attn_drop_ratio=0., drop_path_ratio=0., + representation_size=None, num_classes=1000, name="ViT-B/16"): + super(VisionTransformer, self).__init__(name=name) + + self.img_size=img_size + self.patch_size=patch_size + self.num_classes = num_classes + self.num_heads = num_heads + + self.embed_dim = embed_dim + self.depth = depth + self.qkv_bias = qkv_bias + self.qk_scale = qk_scale + self.drop_ratio = drop_ratio + self.attn_drop_ratio = attn_drop_ratio + self.drop_path_ratio = drop_path_ratio + self.representation_size = representation_size + + # 这里实例化了PatchEmbed类 + self.patch_embed = PatchEmbed(img_size=img_size, patch_size=patch_size, embed_dim=embed_dim) + # 将PatchEmbed类中的num_patches取出来赋值给num_patches + num_patches = self.patch_embed.num_patches + # 这里实例化了ConcatClassTokenAddPosEmbed类 + self.cls_token_pos_embed = ConcatClassTokenAddPosEmbed(embed_dim=embed_dim, + num_patches=num_patches, + name="cls_pos") + + self.pos_drop = layers.Dropout(drop_ratio) + + dpr = np.linspace(0., drop_path_ratio, depth) # stochastic depth decay rule + # 用一个for循环重复Block模块 + # 在用droppath时的drop_path_ratio是由0慢慢递增到我们所指定的drop_path_ratio的 + # 所以我们在构建Block时，这里的drop_path_ratio时变化的，所以用 np.linspace方法创建一个等差数列来初始化drop_path_ratio + self.blocks = [Block(dim=embed_dim, num_heads=num_heads, qkv_bias=qkv_bias, + qk_scale=qk_scale, drop_ratio=drop_ratio, attn_drop_ratio=attn_drop_ratio, + drop_path_ratio=dpr[i], name="encoderblock_{}".format(i)) + for i in range(depth)] + + self.norm = layers.LayerNormalization(epsilon=1e-6, name="encoder_norm") + + # 接下来，如果传入了representation_size，就构建一个全连接层，激活函数为tanh + # 如果没有传入的话，就不做任何操作 + if representation_size: + self.has_logits = True + self.pre_logits = layers.Dense(representation_size, activation="tanh", name="pre_logits") + else: + self.has_logits = False + self.pre_logits = layers.Activation("linear") + + # 定义最后一个全连接层，节点个数就是我们的分类个数num_classes + self.head = layers.Dense(num_classes, name="head", kernel_initializer=initializers.Zeros()) + + def get_config(self): + # 自定义层里面的属性 + config = ( + { + 'img_size': self.img_size, + 'patch_size': self.patch_size, + 'embed_dim': self.embed_dim, + 'depth': self.depth, + 'num_heads': self.num_heads, + 'qkv_bias': self.qkv_bias, + 'qk_scale': self.qk_scale, + 'drop_ratio': self.drop_ratio, + 'attn_drop_ratio': self.attn_drop_ratio, + 'drop_path_ratio': self.drop_path_ratio, + 'representation_size': self.representation_size, + 'num_classes': self.num_classes + } + ) + base_config = super(VisionTransformer, self).get_config() + return dict(list(base_config.items()) + list(config.items())) + + + def call(self, inputs, training=None): + # [B, H, W, C] -> [B, num_patches, embed_dim] + x = self.patch_embed(inputs) # [B, 196, 768] + x = self.cls_token_pos_embed(x) # [B, 197, 768] + x = self.pos_drop(x, training=training) + + for block in self.blocks: + x = block(x, training=training) + + x = self.norm(x) + # 这里是提取class_toke的输出，然后用切片的方式，而刚刚是将class_toke拼接在最前面的 + # 所以这里用切片的方式，去取class_toke的输出，并将它传递给pre_logits + x = self.pre_logits(x[:, 0]) + # 最后传递给head + x = self.head(x) + # 为什么只用class_toke对应的输出，而不用每一个patches对应的输出呢？ + # 可以参考原文bird 网络 + + return x diff --git a/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/__init__.py b/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/__init__.py new file mode 100644 index 0000000..af390d4 --- /dev/null +++ b/TensorFlow_eaxmple/Model_train_test/model/SelfAttention/__init__.py @@ -0,0 +1,8 @@ +#-*- encoding:utf-8 -*- + +''' +@Author : dingjiawen +@Date : 2023/10/23 10:52 +@Usage : +@Desc : +''' \ No newline at end of file