基于卷积历史序列分解混合-长短期记忆网络的锂电池SOC估计OA

Accurate estimation of state of charge(SOC)for lithium-ion batteries is crucial for the energy storage system and electric vehicle energy management.To address the limited SOC estimation accuracy of single neural networks under complex operating conditions,this paper proposes a hybrid estimation model based on a convolutional past decomposable mixing(CPDM)-long short-term memory(LSTM)network.First,the average pooling and a one-dimensional convolutional neural network are used to construct and extract multiscale temporal features from battery data.Second,a CPDM module is applied to perform cross-scale decomposition and mixing to enhance information complementarity.Finally,the enhanced multiscale sequences are fed in parallel into the LSTM network for prediction and the SOC estimation results are obtained by summing the per-scale predictions with equal weights.Experimental results show that the CPDM-LSTM model delivers good SOC estimation performance on public datasets.Under different temperatures and operating conditions,the average root-mean-square error is 0.048 5 and the mean absolute error is 0.037 1,demonstrating strong robustness and generalization of the model.