Unsupervised spatio-temporal state estimation for fine-grained anomaly diagnosis of cyber-physical systems

To reveal the spatio-temporal dependence and evolution mechanisms in cyber-physical system operational states, a fine-grained adaptive multivaviate time series anomaly diagnosis (MAD-Transformer) model was proposed for identifying and diagnosing anomalies in multivariate time series (MTS). MAD-Trans...

Full description

Saved in:
Bibliographic Details
Main Authors: SUN Haili, HUANG Yan, HAN Lansheng, ZHOU Chunjie
Format: Article
Language:zho
Published: Editorial Department of Journal on Communications 2025-07-01
Series:Tongxin xuebao
Subjects:
Online Access:http://www.joconline.com.cn/zh/article/doi/10.11959/j.issn.1000-436x.2025129/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:To reveal the spatio-temporal dependence and evolution mechanisms in cyber-physical system operational states, a fine-grained adaptive multivaviate time series anomaly diagnosis (MAD-Transformer) model was proposed for identifying and diagnosing anomalies in multivariate time series (MTS). MAD-Transformer first constructed temporal state matrixes to characterize and estimate the evolutionary patterns of system states along the time dimension. Secondly, to locate the anomalies, spatial state matrixes were constructed to capture the inter-sensor state correlation. Subsequently, a triple-branch sequence-temporal-spatial attention module was designed to simultaneously capture the sequential, temporal, and spatial dependencies among MTS. Afterwards, three associated alignment loss functions and a reconstruction loss were constructed to jointly optimize the model. The experimental results show that the MAD-Transformer can not only accurately detect and locate the anomaly, but also fine-grained diagnose the duration of the anomaly.
ISSN:1000-436X