Anomaly detection method for edge management devices in energy storage systems

Accurate identification of anomalies in battery clusters is essential to ensuring the safe and stable operation of energy storage stations. In response to the limitations of the cloud-edge-end architecture—namely, excessive data transmission to the cloud, heavy computational load on cloud platforms, and insufficient real-time control—this study proposes an edge-side analytical approach. Voltage data from battery packs are processed at the edge using a boundary point filtering technique based on the Local Outlier Factor (LOF), in combination with Singular Value Decomposition (SVD) and a dynamic grid partitioning scheme. This method significantly improves the computational efficiency of LOF while preserving anomaly detection accuracy. Experimental evaluations conducted using six months of real-world energy storage station operation data demonstrate that the proposed SVD-based dynamic grid LOF boundary filtering method yields superior detection performance on time-series data compared to conventional LOF clustering. • A battery anomaly detection method on edge-terminal side • Improving algorithm efficiency through dimensionality reduction techniques • Adopting a multi-stage detection approach to suit the computational constraints of embedded systems