储能科学与技术2025，Vol.14Issue(10)：3742-3754,13.DOI:10.19799/j.cnki.2095-4239.2025.0301
基于时序模式注意力机制和孤立森林的电池过充热失控预警方法研究
Warning method for battery-overcharge thermal runaway based on a temporal pattern attention mechanism and an isolated forest algorithm
摘要
Abstract
Lithium-ion battery(LIB)thermal runaway proceeds rapidly and violently representing a significant challenging in the field of energy-storage safety,as it can initiate thermal propagation within energy storage systems and lead to losses.To meet the timeliness requirements for early warning of overcharge-induced thermal runaway in LIBs,an early warning method integrating the temporal pattern attention(TPA)mechanism,bidirectional long short-term memory(BiLSTM)network,and isolation forest algorithm is proposed.In this method,the changing patterns of battery-state characteristics across multiple time steps are first captured using TPA,which applies differentiated weighting to focus on the mostvaluable information.Subsequently,the bidirectional neural network structure of BiLSTM is used to extract bidirectional information from battery-characteristic data,thus enhancing the prediction accuracy of the model.Finally,by integrating the isolation forest algorithm,the real battery dataset is used to establish an isolation forest model,which calculates the anomaly scores of battery-state characteristics.Thereafter,the battery states are classified by selecting the optimal anomaly score threshold,enabling early warning of abnormal battery conditions.The experimental results reveal that the proposed method achieves an F1 score of 0.9509 and can provide early warning of abnormal battery conditions 7 s before overcharge-induced thermal runaway;this time is 252 s earlier than that of the temperature threshold method.Overall,the method demonstrated here offers insights into enhancing the accuracy and timeliness of early warning for overcharge-induced thermal runaway in LIBs.关键词
锂离子电池/热失控/预警方法/机制/孤立森林Key words
lithium-ion battery/thermal runaway/warning methods/mechanism/isolation forest分类
信息技术与安全科学引用本文复制引用
李昌豪,曹志成,汪书苹,谢恒,张炜鑫,曹元成..基于时序模式注意力机制和孤立森林的电池过充热失控预警方法研究[J].储能科学与技术,2025,14(10):3742-3754,13.基金项目
国家电网有限公司总部管理科技项目资助(5500-202220118A-1-1-ZN). （5500-202220118A-1-1-ZN）
