Answers are grounded in this paper and cite their sources.
Accurate state-of-charge (SOC) prediction is essential for ensuring the reliable operation of Electric Vehicle (EV) Battery Management Systems (BMS), particularly under varying seasonal and operating conditions. This study proposes a hybrid Bidirectional Long Short-Term Memory–Convolutional Gated Self-Attention (BiLSTM-CGSA) model to improve SOC prediction accuracy and generalization. Specifically, Periodic Encoding (PE) is introduced to incorporate seasonal information into the input features, enhancing the representation of periodic SOC variations. The proposed Convolutional Gated Self-Attention (CGSA) module integrates one-dimensional convolution and gating operations with a Self-Attention Mechanism (SAM), enabling local dynamic feature extraction and adaptive channel-wise feature selection while capturing global temporal dependencies. Experimental validation on real-world driving data and the Urban Dynamometer Driving Schedule (UDDS) at 0 ∘C, 10 ∘C, and 25 ∘C demonstrates that BiLSTM-CGSA achieves superior accuracy and robustness. Specifically, the model achieves a Mean Absolute Error (MAE) of 0.0800; a Mean Squared Error (MSE) of 0.0114; a root Mean Squared Error (RMSE) of 0.1069 on real driving data; and optimal performance at 25 ∘C in UDDS with MAE of 0.0013, MSE of 2.255 × 10−6, and RMSE of 0.0015, representing an average improvement of approximately 77% over six mainstream models. These results indicate that BiLSTM-CGSA effectively captures both global temporal dependencies and local dynamic variations, providing a reliable and efficient approach for intelligent SOC prediction in EV BMS.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: