A hybrid deep learning framework for lithium-ion battery charge estimation under noisy operating conditions

This is the authors' abstract. We don't add key points for this paper.

Accurate state of charge (SOC) estimation is essential for improving the safety, operational reliability, and energy management performance of lithium-ion batteries employed in electric vehicles (EVs) and energy storage systems. However, conventional SOC estimation techniques often exhibit limited robustness under nonlinear operating conditions, sensor noise, battery ageing, and dynamic load variations. To address these challenges, this study proposes a hybrid deep learning model integrating Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks for reliable SOC estimation. The proposed model was evaluated on both synthetically generated battery datasets and the NASA B0005 lithium-ion battery datasets. The Model performance was analysed using mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). Results show that the model achieved an MAE of 1.33%, RMSE of 1.74% and MAPE of 4.04% on the synthetic dataset and an MAE of 0.672%, RMSE of 1.11 and MAPE of 3.15% on held-out NASA B0005. A comparative study with a standalone LSTM, CNN, and Extended Kalman Filter (EKF) has also been performed. Further, an ablation study confirmed that the selected architecture of the model is optimal. Zero-shot evaluation on three additional NASA cells (B0006, B0007, B0018) showed the generalization capacity of the model. Overall, the results demonstrate that the proposed CNN–BiLSTM architecture achieves reliable SOC estimation and indicate its potential applicability in advanced battery management systems for intelligent electric mobility and energy storage applications.