基于深度学习的电动汽车锂电池寿命预测模型

Lithium battery life prediction model for electric vehicles based on hybrid deep learning

摘要:

针对目前电动汽车锂电池剩余使用寿命预测存在预测性能低的问题，提出一种基于混合深度学习的电动汽车锂电池剩余使用寿命预测模型。通过经验模态(EMD)分解将电池数据分解，形成电池容量序列的高频和低频分量；通过使用多层长短时记忆(LTSM)和Elman神经网络学习高频和低频电池容量特征，提取电池容量高层表示；通过叠加规则组合高频和低频预测结果，实现电池剩余使用寿命高精确度预测。实验结果表明，所提混合深度学习检测模型在训练集产生的损失约为7.87%。与支持向量机(SVM)、逻辑回归(LR)、循环神经网络(RNN)和LSTM模型相比，所提混合深度学习模型综合指标性能更优，平均绝对百分比误差(MAPE)仅为1.438%。实验验证了所提模型的有效性及实用性。

Abstract:

In response to the current issue of low prediction performance in the remaining service life of electric vehicle lithium batteries, a hybrid deep learning model for predicting the remaining service life of electric vehicle lithium batteries is proposed. The model employs Empirical Mode Decomposition(EMD) to decompose battery data, forming high-frequency and low-frequency components of the battery capacity sequence. It utilizes Multilayer Long Short-Term Memory(MLSTM) and Elman neural networks to learn high-frequency and low-frequency battery capacity characteristics, extracting high-level representations of battery capacity. It combines high-frequency and low-frequency prediction results through stacking rules to achieve high-precision prediction of the battery's remaining service life. Experimental results show that the loss generated by the proposed hybrid deep learning detection model in the training set is approximately 7.87%. Compared with Support Vector Machine(SVM), Logistic Regression(LR), Recurrent Neural Network(RNN), and LSTM models, the proposed hybrid deep learning model demonstrates superior comprehensive performance indicators, with an Mean Absolute Percentage Error(MAPE) of only 1.438%. The experiments validate the effectiveness and practicality of the proposed model.