State-of-Charge Estimation of Lithium-Ion Batteries Based on the CNN-Bi-LSTM-AM Model Under Low-Temperature Environments

Accurate state-of-charge (SOC) estimation is essential for lithium-ion battery management, especially under low temperatures where traditional methods suffer from noise sensitivity and nonlinear dynamics. In this paper, a hybrid deep learning model integrating a one-dimensional convolutional neural network (1D-CNN), bidirectional long short-term memory (Bi-LSTM), and an attention mechanism (AM) is introduced to enhance SOC estimation accuracy. The 1D-CNN extracts local features from voltage and current sequences, while Bi-LSTM captures bidirectional temporal dependencies, and the AM dynamically emphasizes critical time steps. Experiments conducted on the Panasonic 18650PF dataset at temperatures ranging from -20 to 0 degrees Celsius show that the proposed CNN-Bi-LSTM-AM model achieves a mean absolute error (MAE) of 0.17-0.77% and a root mean square error (RMSE) of 0.33-0.94% under US06 and UDDS driving cycles, outperforming CNN-LSTM and CNN-Bi-LSTM benchmarks. The results demonstrate that the model effectively handles voltage distortion and nonlinearities in low-temperature environments, offering a reliable solution for battery management systems operating under extreme conditions.