基于改进TCN的多元时间序列异常检测算法  

Multivariate time series anomaly detection algorithm based on improved TCN

作　　者：袁安妮 邹春明[2] 王勇 胡津铭 YUAN An-ni;ZOU Chun-ming;WANG Yong;HU Jin-ming(College of Computer Science and Technology,Shanghai University of Electric Power,Shanghai 200120,China;Testing and Certification Center,The Third Research Institute of Ministry of Public Security,Shanghai 200031,China)

机构地区：[1]上海电力大学计算机科学与技术学院,上海200120 [2]公安部第三研究所检测认证中心,上海200031

出　　处：《计算机工程与设计》2025年第2期416-422,共7页Computer Engineering and Design

基　　金：国家重点研发计划基金项目(2021YFF0602701);上海市自然科学基金项目(20ZR1455900)。

摘　　要：为有效解决现有的基于深度学习的多元时间序列异常检测方法中模型对数据要求高,以及捕捉不同时序数据间的相关性差的问题,提出一种从局部与全局信息上改进时序卷积神经网络(temporal convolution network,TCN)的深度学习算法。学习不同时间粒度下数据间的特征依赖关系,在对时序数据特征进行捕捉分析后,通过重构和预测的联合优化实现并行计算数据的异常值。经仿真实验验证,改进后的算法相较其它传统算法,异常检测效果有明显提升。To effectively address the challenges faced by existing multivariate time series anomaly detection methods based on deep learning,such as high data requirements and insufficient capture of correlations between different time series data.A method utilizing local and global information was proposed.A deep learning algorithm based on temporal convolutional networks(TCN)was employed to learn feature dependencies across data with varying time granularities.After capturing and analyzing time series data features,joint optimization of reconstruction and forecasting was performed,enabling parallel computation of anomaly score in the data.Through simulation experiments,it is verified that this improved algorithm significantly enhances the anomaly detection capability compared to traditional methods.

关 键 词：深度学习 时间序列 异常检测 相关性 特征依赖 联合优化 异常值 

分 类 号：TP306.3[自动化与计算机技术—计算机系统结构]