改进EWMA和多通道混合模型的锂电池容量预测  

Lithium Battery Capacity Prediction Using Improved Exponentially Weighted Moving Average and Multi-channel Hybrid Model

作　　者：叶震 李琨[1] 李梦男 吴聪 高宏宇 YE Zhen;LI Kun;LI Mengnan;WU Cong;GAO Hongyu(Faculty of Information Engineering and Automation,Kunming University of Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500

出　　处：《机械科学与技术》2025年第3期513-520,共8页Mechanical Science and Technology for Aerospace Engineering

基　　金：国家自然科学基金项目(82160787)。

摘　　要：针对锂电池在长期循环充放操作后,其剩余可用容量难以准确预测的问题,提出一种改进EWMA(Exponentially weighted moving average)和多通道混合模型的锂电池可用容量预测方法。提取锂电池充电、放电阶段的电压、电流等特性参量,选取其中与容量相关性较高的特征作为健康因子,利用改进EWMA方法对容量数据及健康因子进行滤波,以减少随机短暂的容量回升现象造成的负面影响,将滤波后的健康因子作为特征向量输入到多通道CNN-BiLSTM-SE混合模型中进行预测。采用CALCE实验中心锂电池数据集进行验证,3组锂电池训练预测数据均采用50%的比例划分,预测结果的平均RMSE(Root mean square error,记为ERMS)达到0.014。It is difficult to predict accurately the residual usable capacity of a lithium battery after its long-term cycle charging and discharging operation.Therefore,a prediction method using the improved exponentially weighted moving average(EWMA)and the multi-channel hybrid model is proposed.It extracts the voltage,current and other characteristic parameters of the lithium battery during its charging and discharging stages and selects the features that have a high correlation with the battery's capacity as the health factor.With the improved EWMA method,it filters the capacity data and the health factor,reduces the negative impact caused by the random transient capacity rebound phenomenon and inputs the filtered health factor as the feature vector into the multi-channel CNN-BiLSTM-SE hybrid model to predict the battery capacity.The CALCE experimental center lithium battery datasets were used to verify the prediction.The proportion of 50%was used to divide lithium battery training prediction data into three groups.The average RMSE(Root mean square error,ERMS)value of prediction results reaches 0.014.

关 键 词：EWMA CNN BiLSTM 混合预测模型 电池容量预测 

分 类 号：TM912[电气工程—电力电子与电力传动]