SOC Prediction of Li-Ion Battery Based on EKF and CNN-BiLSTM-Attention

) batteries is critical for ensuring the reliability and safety of commercial and industrial energy storage systems. Deep learning methods have become an important approach for predicting SOC because of their capability to characterize features in an end-to-end manner. However, prediction frameworks that are purely data-driven are prone to performance degradation when they lack physical prior constraints, making it difficult to guarantee the physical consistency and mechanistic interpretability of the prediction results. In this paper, we propose a prediction model that integrates the extended Kalman filtering algorithm (EKF), convolutional neural networks (CNN), bidirectional long short-term memory networks (BiLSTM), and an attention mechanism. First, a second-order resistor-capacitor (RC) circuit model of the battery is established, and SOC is initially estimated using the EKF algorithm. Second, the CNN extracts feature information and passes it into the BiLSTM network, while the attention mechanism assigns weights to enhance the influence of important features. Experimental validation under multiple operating conditions at 10 and 25 °C demonstrates that the proposed method reduces the average estimation error to within 0.5%. Compared with representative models such as CNN-BiLSTM-Attention and CNN-LSTM-Attention, the proposed method maintains lower prediction bias and smaller fluctuation amplitude over the entire time domain. In addition, it shows better generalization capability across different operating conditions and stronger robustness near boundary conditions.