Multivariate Time Series Anomaly Detection Based on Time-Frequency Dynamic Analysis

With the rapid advancement of internet of things and big data technologies, there is a growing demand for accurate and efficient in processing and analysis of extensive time series data in various application scenarios. However, traditional anomaly detection methods ineffective in large-scale, high-dimensional,and dynamically changing time series data. In response to this challenge, this paper proposes a novel time series anomaly detection method that integrates the dynamic graph attention mechanism with time-frequency domain feature analysis. This integration enhances the model's sensitivity to temporal and frequency-related data changes, allowing it to capture intricate spatial dependencies within time series. Moreover, by conducting a comprehensive analysis of time and frequency domain features, the proposed method named TFAD-GAT,which reveals profound patterns and detects abnormal behaviors concealed within the data. Compared with traditional methods, this method improves the accuracy, recall and F1 score of anomaly detection in the SMD dataset, and the recall and F1 score in the SWaT dataset.