WAMA: A Distributed Workload-Aware and Multi-Level Anomaly Detection Framework for Real-Time Smart Grid Device Monitoring

Abstract

Real-time anomaly detection is paramount for ensuring the stable operation of smart grids, but the rapid expansion of sensor networks and increasing complexity of power equipment introduce significant challenges: imbalanced computational loads from heterogeneous data update frequencies, and redundant processing due to exhaustive device-level checks. To address these issues, we propose WAMA, a novel distributed framework for real-time anomaly detection in smart grids. WAMA’s architecture is built upon two key innovations: graph-based workload-aware partitioning (GWAP), which models the power sensing network as a weighted graph to achieve balanced workload distribution while preserving spatial locality; and a multi-level detection (MLD) mechanism, which employs a “trigger-and-refine” cascade across three tiers (partition-unit-device) to progressively refine anomaly identification and significantly reduce computational overhead. Extensive experimental evaluations demonstrate WAMA’s superior performance. Specifically, WAMA’s GWAP improves load balance by a factor of 2 and delivers significantly higher throughput compared to traditional partitioning methods. Furthermore, its MLD mechanism achieves an average throughput an order of magnitude higher than that of the baselines. As a result, WAMA achieves an end-to-end efficiency approximately 26.5 times higher than that of the baselines.