CASTrans unsupervised timing anomaly detection networks

Unsupervised time series anomaly detection has always been a challenging topic. Compared with supervised time series anomaly detection, it can find the abnormal occurrence point more quickly, efficiently and accurately. In this paper we propose a strengthened feature attention module, Channel-Attention Block(CA Block) On the basis of Block, we construct the feature extraction module CABlock-Relu-CABlock (CRC), and design a new self-attention module Channel-Attention Series combined with self-attention Self-Attention(CASSA), and finally built Channel-Attention Series Transformer(CASTrans) with the above two basic modules CRC and CASSA. We have conducted full experiments on the time series anomaly detection datasets PSM, MSL, SMAP and SMD.The experimental results show that, compared with the advanced methods in recent years, most of the indexes of Recall, Precision and F1Score of CASTrans are higher than other SOTA methods.