基于稀疏数据的电动汽车动力电池热故障预警  

Power Battery Thermal Fault Warning of Electric Vehicles Based on Sparse Data

作　　者：张华钦 洪吉超 陈德龙 Zhang Huaqin;Hong Jichao;Chen Delong(School of Mechanical Engineering,University of Science and Technology Beijing,Beijing 100083;Shunde Innovation School,University of Science and Technology Beijing,Foshan 528000;School of Systems Engineering,National University of Defense Technology,Changsha 410073)

机构地区：[1]北京科技大学机械工程学院,北京100083 [2]北京科技大学顺德创新学院,佛山528000 [3]中国人民解放军国防科技大学系统工程学院,长沙410073

出　　处：《汽车技术》2022年第11期24-34,共11页Automobile Technology

基　　金：国家自然科学基金青年基金(52107220);中国博士后科学基金面上项目(2021M690353);佛山市人民政府科技创新专项-产学研合作项目(BK21BE012);北京科技大学青年教师学科交叉研究项目(FRF-IDRY-21-013)。

摘　　要：针对稀疏数据难以准确预警电动汽车动力电池系统热故障和热失控的问题,提出了一种基于长短时记忆网络和迁移学习方法的锂电池热故障预警模型。利用源域密集数据训练该模型,并通过稀疏数据将模型迁移到目标域。在源域训练阶段,利用移动标准差预提取输入数据中的温度相关特征。在目标域训练阶段,提出了放缩指数误差损失函数迁移模型,使故障预警模型自动趋向提取粗糙特征信息,提高温度预测的准确性。试验结果表明,经过迁移学习的神经网络模型能够准确预测电池温度及其变化趋势,对动力电池热故障预警和热失控防控具有重要意义。To address the problem that is difficult for sparse temperature data to perform accurate early warning of thermal failure and thermal runaway for electric vehicle battery systems, a lithium battery thermal fault warning model based on long and short-term memory network and transfer learning is proposed. The model is trained using dense data in the source domain and transferred to the target domain using sparse data. During training phase in the source domain, the moving standard deviation is applied to pre-extract the temperature-related features from the input data. During training phase in the target domain, the loss function transfer model for the scaling exponent errors is proposed to make the fault warning model automatically tend to extract coarse feature information, and improve the accuracy of temperature prediction.The experimental results show that the neural network model with transfer learning can accurately predict the battery temperature and its trends to change, which are important for power battery thermal fault warning and thermal runaway prevention and control.

关 键 词：电动汽车 热故障预警 稀疏数据 迁移学习 移动标准差 

分 类 号：U469.72[机械工程—车辆工程] TM912[交通运输工程—载运工具运用工程]