基于递进预测法的锂电池剩余使用寿命预测  

Remaining service life prediction of lithium battery based on recursive prediction method

作　　者：吴铁洲[1] 刘冉阳 王飞年 汪少夫 梁梦君 WU Tiezhou;LIU Ranyang;WANG Feinain;WANG Shaofu;LIANG Mengjun(School of Electrical and Electronic Engineering,Hubei University of Technology,Wuhan Hubei 430068,China)

机构地区：[1]湖北工业大学电气与电子工程学院,湖北武汉430068

出　　处：《电源技术》2024年第12期2410-2418,共9页Chinese Journal of Power Sources

基　　金：国家自然科学基金项目面上项目(51677058)。

摘　　要：针对锂离子电池剩余使用寿命(remaining useful life,RUL)预测精度低的问题,提出一种递进预测的方法。首先,使用训练集中放电容量和放电电压训练卷积神经网络(convolutional neural netwok,CNN)-长短期记忆网络(long short-term memory neural network,LSTM)模型,并应用于训练集得到预测循环寿命,完成初步预测。其次,使用双指数模型(double exponential model,DEM)从训练集中辨识与预测循环寿命最接近的电池参数,并作为均值函数输入高斯过程回归(Gaussian process regression,GPR)模型。最后,使用测试集的电池循环数和电池容量训练GPR模型,并用于RUL预测。实验结果表明,本方法在99%的置信水平下,平均绝对百分比误差在6%以内,准确率百分比在90%以上,证明了所提方法的可靠性。Aiming at the problem of low prediction accuracy of the remaining useful life(RUL)of lithium-ion batteries,this paper proposed a progressive prediction method.Firstly,the convolutional neural network(CNN)-long short-term memory neural network(LSTM)model was trained using the discharge capacity and discharge voltage in the training set,and LSTM model was used to the training set to get the predicted cycle life and complete the preliminary prediction.Secondly,the double exponential model(DEM)was used to identify the battery parameters in the training set that are closest to the predicted cycle life and input them as the mean function into GPR model.Finally,GPR model was trained using the number of battery cycles and battery capacity from the test set and used for RUL prediction.The experimental results show that at a 99%confidence level,the mean absolute percentage error of this method is within 6%and the accuracy percentage is above 90%,which proves the reliability of the proposed method.

关 键 词：锂电池 剩余寿命预测 卷积神经网络 长短时神经网络 高斯过程回归 

分 类 号：TM912.9[电气工程—电力电子与电力传动]