Bridging Edge and Cloud: A Knowledge-Enhanced Framework for Efficient Time Series Anomaly Detection

Multivariate time series anomaly detection (MTSAD) plays a crucial role in monitoring system health and ensuring operational reliability across various domains. While deep learning models have shown promising results in this field, deploying these computationally intensive models on resource-constrained edge devices remains challenging, particularly when considering the need for continuous model adaptation to evolving data patterns. We introduce the RefinedEdge framework, explicitly developed to enable effective deployment of multivariate time series anomaly detection within resource-constrained edge computing environments. The RefinedEdge framework utilizes three key strategies: Aggregated Compression, Knowledge Refinement, and Reciprocal Edge-Cloud Learning. These strategies collectively tackle prominent challenges such as maintaining high accuracy despite significant model compression and ensuring seamless synchronization and continuous adaptation across cloud and edge systems. By reducing the complexity of models without sacrificing performance, and by adapting anomaly detection to the dynamic conditions typical of edge computing, the RefinedEdge framework validates its significant enhancements in detection performance through empirical testing on real-world datasets. This makes it a practical solution for real-time applications in sectors such as smart manufacturing.