Multiscale Wavelet Graph AutoEncoder for Multivariate Time-Series Anomaly Detection

In industrial facilities or IT systems, there are lots of multivariate time series generated from various metrics. Anomaly detection in multivariate time series is of great importance in applications such as fault diagnosis and root cause discovery. Recently, some unsupervised methods have made great progress in this task, especially the reconstruction architecture of autoencoders (AEs), learning normal distribution, and producing a significant error for anomalies. Although AEs can reconstruct subtle abnormal patterns well with the powerful generalization ability, it also leads to a high false negative. Moreover, these AE-based models ignore the dependence among variables at different time scales. In this article, we propose a novel anomaly detection framework named Multiscale wavElet Graph AE (MEGA). The main idea is to integrate discrete wavelet transform (DWT) into AE to decompose multivariate time series into multifrequency components and then reconstruct them, thereby highlighting various anomalies in specific frequency bands. Meanwhile, for anomalies caused by the changes of intervariable dependence, we introduce a dynamic graph module to capture such dependence on the decomposed multiscale frequency components. Experiments have been carried out on three public multivariate time-series anomaly detection datasets. The results demonstrate that the MEGA outperforms the state-of-the-art baselines.

Automated summary

In this article, a novel anomaly detection framework named MEGA is proposed. It integrates discrete wavelet transform (DWT) into autoencoders (AE) to decompose multivariate time series into multifrequency components and then reconstruct them, highlighting various anomalies in specific frequency bands. A dynamic graph module is introduced to capture anomalies caused by changes in intervariable dependence on the decomposed multiscale frequency components. Experiments show that MEGA outperforms existing baselines.