基于优化Elman神经网络的锂电池容量预测

Capacity prediction of lithium battery based on optimized Elman neural network

锂离子电池容量的精确预测有利于提升电池的使用安全和避免电池滥用,受其复杂的内部电化学反应和外部使用条件影响,对其老化后的容量精确预测一直是电池管理系统的难点之一。为实现锂电池全服役周期内容量的高效准确预测,本文提出了一种基于遗传算法优化的Elman神经网络(GA-Elman)电池容量预测模型。首先选择电池不同循环下的放电容量增量、内阻以及温度数据作为有效表征电池老化和容量衰减规律的特征量,其次运用主成分分析算法对特征量进行降维以降低训练量数据维度,然后基于Elman神经网络构建电池容量预测模型,并引入遗传算法优化Elman神经网络的权值和阈值,实现对电池容量的高效精确预测,最后在不同电池上对该模型进行了验证。验证结果表明:与传统Elman神经网络和长短期记忆神经网络(long and short term memory neural network,LSTM NN)预测模型相比,GA-Elman神经网络预测模型有更好的预测精度和更高的运算效率。在不同电池上该模型预测结果的最大平均绝对误差为0.92%,最大均方根误差为1.02%,最小拟合系数为0.9679,表明该模型可以精确预测锂电池衰退过程中的容量并且对不同电池有较强的适应性。

Accurate prediction of lithium-ion battery capacity is advantageous for enhancing battery safety and avoiding battery abuse.An accurate capacity prediction has always been a challenge in battery management systems due to the effect of complicated internal electrochemical reactions and external variables.To achieve an effective and accurate prediction of lithium battery capacity in the full-service life,this study proposed an Elman neural network battery capacity prediction model optimized by a genetic algorithm.To begin,the typical characteristics of discharge capacity growth,internal resistance,and temperature data collected over various battery cycles were chosen to adequately depict the law of battery aging and capacity degradation.Secondly,the principal component analysis algorithm is used to reduce the dimension of the characteristic quantities to reduce the data dimension of the training quantity.The Elman neural network is then used to build the battery capacity prediction model.To ensure efficient and accurate prediction of battery capacity,a genetic algorithm is used to adjust the weights and thresholds of the Elman neural network.Finally,the model was verified on different batteries.The verification findings reveal that the GAElman neural network prediction model is more accurate and efficient than the classic Elman neural network and LSTM neural network prediction models.The maximum mean absolute error,maximum root mean square error,and minimum fitting coefficient of the model is 0.92%,1.02%,and 0.9679,respectively,for various batteries,showing that the model can accurately predict the capacity of lithium battery in the process of decay and has high adaptability to different batteries.