WAGAN:基于小波变换和注意力机制的工控传感器数据异常检测方法  

WAGAN:Industrial Control Sensor Data Anomaly Detection Method Based on Wavelet Transform and Attention Mechanism

作　　者：马标 贾俊铖[1] 董国柱 章红 陆武民 MA Biao;JIA Jun-cheng;DONG Guo-zhu;ZHANG Hong;LU Wu-min(School of Computer Science and Technology,Soochow University,Suzhou 215006,China;Zhejiang Supcon Technology Co.,Ltd.,Hangzhou 310058,China;School of Electronic Andinformation Engineering,Soochow University,Suzhou 215006,China)

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006 [2]浙江中控技术股份有限公司,杭州310058 [3]苏州大学电子信息学院,江苏苏州215006

出　　处：《小型微型计算机系统》2023年第1期168-176,共9页Journal of Chinese Computer Systems

基　　金：中国博士后科学基金项目(2017M611905)资助;江苏高校优势学科建设工程项目(PAPD)资助。

摘　　要：针对隐藏攻击意图的入侵行为,通过现场设备传感器实时数据反映工业控制系统运行情况,充分利用工控数据高周期性特点,提出WAGAN(Wavelet Attention Generative Adversarial Networks)的工控传感器数值异常检测方法.此方法使用多级离散小波变换分解重组的方式去除噪声并增强数据特征.为了有效提取数据的有效特征,在WAGAN模型中引入了注意力机制,并使用多层LSTM(Long Short-Term Memory)网络学习数据的潜在关联性.为了提高模型准确性,使用生成器的重构误差与判别器误差的权重和来判断异常.实验结果表明,此方法相比于现有的异常检测方法具有更高的异常检出率.Aiming at the intrusion behavior with hidden attack intentions, real-time data from field device sensors reflect the operation of industrial control systems, making full use of the high periodicity of industrial control data, and propose WAGAN(Wavelet Attention Generative Adversarial Networks) industrial control sensor numerical anomaly detection method.This method uses multi-level discrete wavelet transform decomposition and recombination to remove noise and enhance data features.In order to effectively extract the effective features of the data, attention mechanism is introduced into the WAGAN model, and a multi-layer LSTM(Long Short-Term Memory) network is used to learn the potential relevance of the data.In order to improve the accuracy of the model, the weighted sum of the reconstruction error of the generator and the error of the discriminator is used to judge the anomaly.Experimental results show that this method has a higher anomaly detection rate than existing anomaly detection methods.

关 键 词：工业控制系统 传感器 异常检测 周期性 

分 类 号：TN751[电子电信—电路与系统]