Pre-training Enhanced Transformer for multivariate time series anomaly detection

Chao Wang, Haochen Shi, Jie Hu, Xin Yang, Junbo Zhang, Shengdong Du, Tianrui Li

Abstract: Highlights•Propose a novel framework for anomaly detection based on Transformer.•Design a pre-training model to generate representations with rich feature patterns.•Utilize a pre-training model to enhance the reconstruction ability of Transformer.•Introduce a universal anomaly knowledge to amplify discrepancies within time series.