基于t-SNE降维和KNN算法的波浪传感器故障诊断方法  

Wave sensor fault diagnosis method based on t-SNE reduction and KNN algorithm

作　　者：邰朋 宋苗苗 王波 陈世哲 付晓 扈威 高赛玉 程凯宇 郑珊珊 焦梓轩 王龙飞 TAI Peng;SONG Miaomiao;WANG Bo;CHEN Shizhe;FU Xiao;HU Wei;GAO Saiyu;CHENG Kaiyu;ZHENG Shanshan;JIAO Zixuan;WANG Longfei(College of Marine Technology and Science,Qilu University of Technology(Shandong Academy of Sciences),Qingdao 266300,China;Institute of Marine Instrumentation,Qilu University of Technology(Shandong Academy of Sciences),Qingdao 266300,China)

机构地区：[1]齐鲁工业大学(山东省科学院)海洋技术科学学院,山东青岛266300 [2]齐鲁工业大学(山东省科学院)海洋仪器仪表研究所,山东青岛266300

出　　处：《山东科学》2023年第4期1-9,共9页Shandong Science

基　　金：青岛市自然科学基金(23-2-1-159-zyyd-jch);国家自然科学基金(50875132,60573172);青岛市科技惠民项目(22-3-7-cspz-16-nsh)。

摘　　要：针对波浪传感器故障诊断困难、故障类型无法识别、诊断耗时长的问题,提出一种基于小波包分解、降维与k-近邻算法(k-nearest neighbor algorithm,KNN)分类网络的波浪传感器故障诊断方法。首先将原始信号进行标准差标准化处理,然后对标准化后的数据进行小波包3层分解,将分解后的第3层8个频带上的数据进行归一化处理,作为提取的特征向量,采用t-SNE降维算法对特征数据进行降维,最后将降维后的特征数据输入到KNN分类网络中进行故障分类检测。实验结果表明,该方法能够提高波浪传感器故障诊断的准确度和诊断速度,对正常状态和6种故障状态的诊断准确率能够达到93.55%。∶This study proposes an efficient wave sensor fault diagnosis method based on wavelet packet decomposition,dimension reduction,and k-nearest neighbor algorithm(KNN)classification network to address the difficulty of wave sensor fault diagnosis,unidentifiable fault types,and time-consuming diagnosis.First,the standard deviation of the original signal is normalized.The normalized data is then subjected to a three-layer wavelet packet decomposition.The extracted feature vectors represent normalized data from the eight bands on layer 3.The second step involves using the t-distributed stochastic neighbor embedding(t-SNE)algorithm to reduce the dimension of the feature data.Finally,the dimension-reduced feature data is input into the KNN classification network for fault classification and detection.Experimental results show that the proposed method can improve the accuracy and diagnosis speed of the wave sensor fault diagnosis,with a diagnosis accuracy of up to 93.55%for normal and six faulty conditions.

关 键 词：波浪传感器 小波包分解 t-SNE非线性降维算法 K-近邻算法 故障诊断 

分 类 号：TP83[自动化与计算机技术—检测技术与自动化装置]