基于深度孤立森林的锂离子电池短路故障诊断方法

Short-circuit faults in lithium-ion batteries (LIBs) pose serious safety risks, including thermal runaway, making timely detection critical for the safe operation of battery systems, especially in new energy power systems with large-scale battery integration. However, diagnosing these faults remains challenging, particularly when voltage fluctuations stay within safety thresholds. This paper proposes a novel method for minor short-circuit (MSC) fault diagnosis in LIBs, based on voltage phase-plane feature extraction and a deep isolation forest algorithm. Voltage phase-plane point sequences are constructed from cell voltage and its first-order difference, from which fault features—2-D correlation coefficient and improved Fréchet distance—are extracted and fed into a deep isolation forest model. By integrating a deep neural network, the model enhances anomaly detection for nonlinear data, enabling both fault detection and localization. Experimental results show that the proposed method achieves real-time detection of MSC faults even when the voltage fluctuation is within the threshold range. The method demonstrates strong robustness and generalization across diverse operating scenarios, crucial for the dynamic environments of new energy power systems. A fault detection rate (FDR) of 96% and a fault misdetection rate (FMR) of 3% are achieved, significantly outperforming traditional approaches such as sample entropy, correlation coefficient analysis, and conventional isolation forest algorithms.