Feature-Aligned and Class-Aware Incremental Learning for Online Partial Discharge Monitoring

The development of a reliable and accurate online condition monitoring system for detecting partial discharges (PD) in electrical switchgears is critical for ensuring the reliability and safety of high-voltage equipment. As the demand for scalable and centralized monitoring grows, existing methods face new challenges in adaptability and performance. In this study, we propose a feature-aligned, class-aware incremental learning framework designed specifically for online PD monitoring in switchgear units distributed across multiple substations. The system aggregates data streams from these distributed sites periodically to facilitate centralized model updates and knowledge refinement. To address the challenges of efficient model updates in the face of increasingly dynamic datasets, as well as the need for rapid decision-making and bandwidth efficiency at the edge, we propose a spatial alignment-based knowledge distillation approach that transfers knowledge from the cloud model to lightweight models deployed on edge-computing devices. This edge–cloud collaborative architecture ensures fast, accurate, and context-aware PD detection at the front side. We organize the captured PD data from substations to assess the efficacy of the proposed method. Experimental results on the PD datasets demonstrate the superior performance of the proposed framework compared to existing methods.