SOC estimation for lithium batteries using a CNN-attention-LSTM model

In order to improve the accuracy of state of charge (SOC) estimation for lithium batteries, a SOC prediction model based on CNN-Attention-LSTM is proposed. In this model, convolutional neural network (CNN) is used to extract the timing characteristics of battery voltage, current and temperature. Then, through the attention mechanism, dynamic weighting is performed to enhance the learning of key features. The dependencies in time-series data are then captured using a Long Short-Term Memory (LSTM). Finally, SOC parameters are output at the regression layer. In order to improve the efficiency and accuracy of model training, this paper introduces the Coronate Porcupine Optimization (CPO) to optimize the network hyperparameters. In this paper, SOC estimation experiments are performed under different temperatures, initial currents and datasets. The experimental results show that the MAE of the model is 0.84 % and 1.06 % at 50 °C and 0 °C, respectively. At 100 % and 60 % initial electricity, the MAE of the model is 1.09 % and 1.15 %, respectively, and the SOC estimation accuracy is significantly better than that of the LSTM and CNN-LSTM models. Furthermore, the model in this paper has high SOC estimation accuracy on different data sets, and the model shows good robustness and generalization. • CNN-Attention-LSTM model significantly improves lithium battery SOC estimation accuracy. • Innovatively introduces Crested Porcupine Optimizer for automatic hyperparameter optimization. • Multi-condition tests demonstrate superior model generalization capability. • Achieves notably lower SOC estimation error rates than traditional LSTM methods.