基于早期老化与迁移学习的电池寿命预测研究  

Research on Battery Life Prediction Based on Early Aging and Transfer Learning

作　　者：杨天 沈进冉 官亦标 高飞 姜添 刘庆 YANG Tian;SHEN Jinran;GUAN Yibiao;GAO Fei;JIANG Tian;LIU Qing(China Electric Power Research Institute,Beijing 100192,China)

机构地区：[1]中国电力科学研究院有限公司,北京100192

出　　处：《湖南电力》2025年第1期1-7,共7页Hunan Electric Power

基　　金：国家电网有限公司科技项目(5419-202323785A-3-8-KJ)。

摘　　要：针对储能电池寿命预测可靠性不足、泛化性不佳的问题,提出一种基于LSTM的储能电池寿命预测方法。预测模型仅采用电池前150个循环的早期老化数据作为输入,通过滑动窗口法训练模型,实现了储能电池寿命的精准预测,1000个循环内均方根误差为0.11%。同时开发迁移学习模块,对于新型号储能电池,采用第150次循环至250次循环对基础模型进行迁移修正,将电池寿命预测均方根误差由0.50%降至0.10%,提升了寿命预测模型的泛化性,为寿命预测技术在储能电池生产及检测评价领域的实际应用提供一种解决方法。To address the problems of insufficient reliability and poor generalization in life prediction of energy storage batteries,a large-capacity energy storage battery remaining life prediction method based on LSTM is proposed,which only adopts the early aging data of the batteryfrom the first 150 cycles as the inputand realizes accurate prediction of the remaining life of energy storage batteries through the model training method of the sliding-window method.This approach achieves a remarkable root mean square error(RMSE)of 0.11%over a full national standard testing period of 1000 cycles.Additionally,a transfer learning module is developedand for the new model of energy storage batteries,the 150th to 250th cycles are used for the migration correction of the basic model,which reduces RMSE of the battery life prediction from 0.50%to 0.10%.This advancement enhances the generalization of the life prediction model,providing a solution for the practical application of the life prediction technology in the field of energy storage battery production,testing and evaluation.

关 键 词：储能电池 寿命预测 迁移学习 早期老化数据 

分 类 号：TM911[电气工程—电力电子与电力传动]