MultiCon-AD: Integrated Contrastive Learning for Multivariate Time Series Anomaly Detection

Time series anomaly detection is a complex task due to the intricate temporal dependencies and scarcity of labeled data. Although contrastive learning methods have been developed, they are often designed for univariate time series, which restricts their effectiveness in handling the complexities of multivariate time series. This paper presents MultiCon-AD, an innovative contrastive learning framework specifically designed for multivariate time series anomaly detection. Our model provides a comprehensive analysis from four key perspectives: Entity-wise, Channel-wise, Contextual-wise, and Transformation-wise. By employing this multi-faceted approach, MultiCon-AD effectively captures data consistency in a self-supervised manner, allowing for more accurate identification of anomalies in complex time series data. The experimental results demonstrate that MultiCon-AD achieves state-of-the-art performance across various real-world multivariate time series datasets.