基于改进自编码模型的工业时序数据异常检测  

Industrial Time Series Data Anomaly Detection Based on Improved Autoencoder Model

作　　者：肖鉴涛 廖光忠[2] XIAO Jiantao;LIAO Guangzhong(School of Computer Science and Technology,Wuhan University of Science and Technology,Wuhan 430065;Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System,Wuhan University of Science and Technology,Wuhan 430065)

机构地区：[1]武汉科技大学计算机科学与技术学院,武汉430065 [2]武汉科技大学智能信息处理与实时工业系统湖北省重点实验室,武汉430065

出　　处：《计算机与数字工程》2023年第11期2627-2632,2670,共7页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61502359)资助。

摘　　要：针对传统异常检测模型在处理工业多维时序数据时特征提取不充分、抗干扰能力弱等问题,提出一种改进的自编码模型,有效结合了门控循环网络的时序信息记忆能力和收缩自编码器的鲁棒特征提取能力,能够同时捕获不同特征变量之间的非线性相关性和单个变量自身的时序相关性。采取半监督学习异常检测方法,使用正常数据训练模型收敛,并根据待检测数据输入模型后计算出的异常得分来判定异常样本。基于真实的工业传感器数据进行实验后的结果表明,该方法有效提高了异常检测的准确度和可靠性。Aiming at the problem of inadequate feature extraction and weak anti-interference capability of traditional anomaly detection models in processing the multivariate industrial time series data,an improved autoencoder model is proposed,which effec⁃tively combines the time series information memory capability of gated recurrent network and the robust feature extraction capability of contractive autoencoder,and is succeeded in capturing both the nonlinear correlation between different feature variables and the time series correlation of individual variables themselves.Based on the semi-supervised learning anomaly detection method,the model is trained to be convergent with normal data,and the anomalous samples are determined according to the abnormality scores calculated after the data to be detected is input to the trained model.Experimental results using the real industrial sensor data indi⁃cate that the accuracy and reliability of anomaly detection are effectively improved by the proposed method.

关 键 词：异常检测 时序数据 半监督学习 收缩自编码器 门控循环网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]