基于CNN-GRU神经网络的锂电池SOH估计与RUL预测  

SOH Estimation and RUL Prediction of Lithium Battery Based on CNN-GRU Neural Networks

作　　者：辛付宇 邢丽坤[1] 刘笑 XIN Fuyu;XING Likun;LIU Xiao(College of Electrical and Information Engineering, Anhui University of Science and Technology)

机构地区：[1]安徽理工大学电气与信息工程学院

出　　处：《上海节能》2024年第5期819-826,共8页Shanghai Energy Saving

摘　　要：随着近几年新能源汽车市场的蓬勃发展,消费者对锂电池的电池性能和储能系统的整体要求逐步提升。锂电池作为新能源汽车的重要组成部分,对新能源汽车品牌的经济性能具有重要影响。针对锂电池健康状态(State Of Health,SOH)估计与剩余有效工作时间(Remaining Useful Life,RUL)预测无法直接测量,为了攻克在线准确测量的难题,提出基于卷积神经网络-门控循环单元(Convolutional Neural Network-Gated Recurrent Unit,CNN-GRU)的锂电池SOH估计与RUL预测方法。运用Python编程语言在TensorFlow框架下搭建CNN-GRU神经网络,利用GRU长时间记忆能力与CNN避免了对数据的复杂前期预处理,采用NASA开放实验数据测试,经过实验结果对比,基于CNN-GRU神经网络的估算模型相对于BP、CNN、GRU单独神经网络模型拥有更高的计算精度,以及更稳定的预测结果。With the rapid development of the new energy vehicle market in recent years,the overall requirements of consumers for the battery performance and energy storage system of lithium batteries have gradually increased.As an important part of new energy vehicles,lithium battery has an important impact on the economic performance of new energy vehicle brands.The State of Health(SOH) estimation and Remaining Useful Life(RUL) prediction of Li-ion batteries cannot be measured directly,and in order to overcome the problem of online accurate measurement,a Convolutional Neural Network-Gated Recurrent Unit(CNN-GRU) is proposed.Built upon the TensorFlow framework and Python programming language,leveraging the long-term memory capabilities of GRU and CNN to avoid the complex pre-processing of data,and using NASA open experimental data test,after the comparison of experimental results,the CNN-GRU-based estimation model demonstrates higher computational efficiency and accuracy with more stable prediction results over traditional BP,CNN,and GRU models.

关 键 词：锂电池 卷积神经网络 门控循环单元 健康状态 剩余有效工作时间 

分 类 号：TM912[电气工程—电力电子与电力传动] TP183[自动化与计算机技术—控制理论与控制工程]