基于改进Shapley-ResNet储能电站锂电池火灾预警模型  

Lithium Battery Fire Warning Model Based on Improved Shapley-ResNet Energy Storage Power Station

作　　者：范旭明 张一航 杨欢红 梁勋萍 江世进 张赛凡 彭涛 Fan Xuming;Zhang Yihang;Yang Huanhong;Liang Xunping;Jiang Shijin;Zhang Saifan;Peng Tao(Jinhua Power Supply Branch,State Grid Zhejiang Electric Power Co.,Ltd.,Jinhua Zhejiang 321002,China;Shanghai University of Electric Power,Shanghai 200090,China)

机构地区：[1]国网浙江省电力有限公司金华供电公司,浙江金华321002 [2]上海电力大学,上海200090

出　　处：《电气自动化》2025年第2期87-89,共3页Electrical Automation

基　　金：国家自然科学基金项目(51777119);国网浙江省电力有限公司科技项目(BD2023JH-KXXM005)。

摘　　要：为提升储能电站锂电池火灾早期预警效率,提出了基于E-Shapley-ResNet的火灾预警模型。首先,构建了储能电站锂电池火灾预警指标体系,涵盖了结构缺陷类、热失控类和早期火灾类,共19项指标;然后利用E-Shapley方法进行指标体系的特征降维,计算各指标在预测模型中的敏感性;最后利用ResNet18算法建立了储能电站火灾预警模型,通过实例对上述模型的可行性进行了验证。结果表明,与传统机器学习算法相比,E-Shapley-ResNet18算法在准确率、召回率和AUC值方面表现较好,尤其是对燃火预测的准确性有显著优势。To improve the early warning efficiency of lithium battery fires in energy storage power stations,a fire warning model based on E-Shapley ResNet was proposed.Firstly,a fire warning index system for lithium batteries in energy storage power stations was established,covering 19 indicators including structural defects,thermal runaway,and early fires;then,the E-Shapley method was used to reduce the feature dimensionality of the indicator system and calculate the sensitivity of each indicator in the prediction model;finally,a fire warning model for energy storage power stations was established using the ResNet18 algorithm,and the feasibility of the above model was verified through examples.The results show that compared with traditional machine learning algorithms,the E-Shapley-ResNet18 algorithm performs better in accuracy,recall,and AUC values,especially in predicting smoldering fires,with significant advantages in accuracy.

关 键 词：储能电站 锂电池 火灾预警 ResNet18算法 Shapley理论 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]