用于电类实验测量数据异常检测的MVAE神经网络  

MVAE Neural Network for Anomaly Detection of Electrical Experimental Measurement Data

作　　者：申赞伟 刘彦博 杨柳[1] 曹淋涵 熊英杰 张峰[1] SHEN Zanwei;LIU Yanbo;YANG Liu;CAO Linhan;XIONG Yingjie;ZHANG Feng(School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《实验室研究与探索》2025年第1期24-29,共6页Research and Exploration In Laboratory

基　　金：教育部—阿里云产学合作协同育人项目(220600643165804);教育部供需对接就业育人项目(20220106373,20220102556,20220104444);上海交通大学决策咨询课题(JCZXSJB2022-08);上海交通大学教学发展基金项目(CTLD22J 0066)。

摘　　要：电类实验课中测量数据对错的人工评判降低了课堂教学的效率和质量。为此,提出一种改进变分自编码器(MVAE)神经网络,利用均值漂移方法得到实验测量数据的隐变量均值;以实验测量数据的训练集样本为输入,基于反向传播算法对MVAE参数进行训练,得到实验测量数据的隐变量标准差和隐变量的正态分布。若待判决的测试样本编码后的数据位于隐变量正态分布的2个标准差范围外,则该样本为异常数据,即错误测量数据。研究结果表明,MVAE模型不仅提高了学习效率,而且提高了异常检测判别的准确率。The manually judging the right or wrong data in electrical experiment courses reduces the efficiency and quality of classroom teaching.To this end,an improved variational autoencoder(MVAE)neural network is proposed.The mean shift method is used to obtain the mean value of the hidden variable of the experimental measurement data.The training samples of the experimental measurement data are used as input to train the parameters of MVAE based on back-propagation algorithm.The standard deviation of the hidden variable of the experimental measurement data is finally obtained,thereby obtaining the normal distribution of the hidden variable.If the encoded data of the test sample to be judged is outside the range of 2 standard deviations of the normal distribution of the hidden variable,the sample is abnormal,that is an incorrectly measured datum.The research result shows that the MVAE model not only improves the learning efficiency,but also improves the accuracy of anomaly detection.

关 键 词：电类实验课程 变分自编码器 神经网络 异常检测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]