基于WGAN-Attention的电厂多变量时间序列异常检测  

Multivariate time series anomaly detection for power plants based on WGAN-Attention

作　　者：戚奉彪 李海广[1] 张超[2] QI Fengbiao;LI Haiguang;ZHANG Chao(School of Energy and Environment,Inner Mongolia University of Science&Technology,Baotou 014010,China;School of Mechanical Engineering,Inner Mongolia University of Science&Technology,Baotou 014010,China)

机构地区：[1]内蒙古科技大学能源与环境学院,内蒙古包头014010 [2]内蒙古科技大学机械工程学院,内蒙古包头014010

出　　处：《内蒙古科技大学学报》2025年第1期86-91,共6页Journal of Inner Mongolia University of Science and Technology

基　　金：国家自然科学基金地区项目(51965052);内蒙古自治区自然科学基金(2019LH05012)。

摘　　要：通过将电厂中的多变量时间序列数据转化为特征图来表征变量之间的相关性信息,将特征图输入至“Wasserstein GAN(WGAN)”中,通过卷积提取特征,在WGAN中添加注意力机制来分配各个特征的权重,将重构的特征图与原始特征图进行对比,通过损失函数来量化分析设置阈值,进行异常检测。实验表明:此方法可以在多变量数据中准确检测出异常变量并进行定位。In this paper,we characterize the correlation information between variables by converting multivariate time series data in a power plant into a feature map,input the feature map into a Wasserstein GAN(WGAN),extract features by convolution,add an attention mechanism to the WGAN to assign the weights of each feature,compare the reconstructed feature map with the original feature map,and quantify the analysis by loss function Set the threshold value for anomaly detection.Experiments show that this method can accurately detect anomalous variables in multivariate data and locate them.

关 键 词：多变量时间序列 异常检测 生成对抗式网络 注意力机制 电厂 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]