Scaling Up Online Model-Based Anomaly Detection Methods for Use in Large Battery Stacks: Addressing Challenges in Online Monitoring of Battery Systems

The online monitoring of battery energy storage systems (ESSs) is crucial to detecting sensor anomalies that could have detrimental impacts on the system’s safety, reliability, and availability. Model-based anomaly detection mechanisms could be integrated into battery management systems (BMSs) with modeling and estimation capabilities; some high-voltage BMSs are capable of controlling up to 36 battery cells simultaneously. However, previous model-based anomaly detection mechanisms have been tested only on small stacks of batteries or single cells. In this article, four model-based anomaly detection methods were applied to a simulated stack of 12 and 36 series-connected battery cells to emulate a high-voltage BMS anomaly detector. The detectors’ false positive rate, detection accuracy, and computational burden were compared, and the Cumulative Sum (CUSUM) algorithm was found to marginally outperform the other three anomaly detectors with a false positive rate of 0%, a detection accuracy of 99.6%, and a maximum computational time of 0.0162 s and 0.0909 s per iteration for the 12-cell and 36-cell stack, respectively.