BiLSTM-Based Anomaly Detection in Multivariate Time Series with Attention Mechanism and Dual Analysis

Anomaly detection in multivariate time series is of significance in industrial equipment fault detection, network security, etc. In light of the arrival of big data, the temporal dependency of the data of multivariate time series features and the interaction between variables have become increasingly complex, and anomaly detection has turned into a formidable challenge. In response to this challenge, this paper presents the BiL-AT method for multivariate time series anomaly detection, which combines bidirectional long short-term memory (BiLSTM) and attention mechanism (AT) to perform dual analysis of prediction and reconstruction on the model. This integration can not only capture bidirectional time series data and improve the scope of captured information, but also extract data with key information through the attention mechanism. At the same time, the dual analysis of reconstruction and prediction, as well as the automatic search for the best anomaly threshold, further calculate the outlier value. This method has obvious advantages in dealing with long-term dependencies, multivariate complex interactions, dynamic and non-stationary data, and improves the accuracy, robustness and real-time performance of anomaly detection.