基于深度降噪自编码神经网络的SCADA系统异常检测  

Deep denoising autoencoder neural network based anomaly detection for SCADA systems

作　　者：满雯妍 李红娇 MAN Wen-yan;LI Hong-jiao(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200090

出　　处：《计算机工程与设计》2023年第7期1977-1984,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61403247、61702321)。

摘　　要：针对SCADA系统数据特征提取难度大导致异常检测准确率低、漏报率高的问题,提出一种基于深度降噪自编码神经网络的SCADA系统异常检测方法。采用降噪自编码器,将前一个自编码器的输出作为下一个自编码器的输入,连接构建深度神经网络结构加强特征学习能力,为优化网络参数,改进重构误差函数,提高模型的重构能力。使用SCADA系统数据集测试所提方法,实验结果表明,与其它异常检测方法相比,该方法在保证较高准确率的同时能有效降低漏报率。Aiming at the problems of low anomaly detection accuracy and high failure rate caused by the difficulty of SCADA systems data feature extraction,an anomaly detection method for SCADA systems based on deep denoising autoencoder neural network(DDAENN)was proposed.The denoising autoencoder was adopted,and the output of the previous autoencoder was taken as the input of the next autoencoder,the deep neural network structure was connected to strengthen the feature learning ability.Meanwhile,to optimize the network parameters,the reconstruction error function was improved to improve the reconstruction ability of the model.Experimental results show that compared with other anomaly detection methods,the proposed method can ensure high accuracy and effectively reduce the false alarm rate.

关 键 词：数据采集与监控系统 异常检测 降噪自编码器 神经网络 深度学习 特征学习 重构误差 

分 类 号：TP393.08[自动化与计算机技术—计算机应用技术] TP183[自动化与计算机技术—计算机科学与技术]