Mechanical stress-based state-of-charge estimation for lithium-ion batteries via deep learning techniques

Owing to the wide voltage platform and severe polarization of lithium-ion batteries (LIBs), traditional methods relying on voltage characteristics cannot accurately estimate the state of charge (SOC), particularly under dynamic conditions. To address this issue, we propose a new PLO-TCNUltra-SE model for SOC estimation, incorporating mechanical stress data alongside conventional parameters such as voltage. A comprehensive evaluation is conducted through feature extraction on the basis of mutual information, Pearson's correlation coefficient, and extreme gradient boosting (XGBoost) algorithms. The model combines an improved temporal convolutional network (TCN) with a squeeze-and-excitation (SE) attention mechanism to capture long-term dependencies and key time steps. Hyperparameter tuning is performed by applying the polar light optimization (PLO) algorithm to adapt the model to varying battery characteristics at different temperatures. Finally, experimental validation is performed on one lithium and two odium battery datasets. When compared with that of the conventional bidirectional gated recurrent unit (BiGRU), long short-term memory (LSTM) and latest Mamba models, the proposed model demonstrates strong performance across all 3 datasets, with root mean square error (RMSE), mean absolute error (MAE) and maximum absolute error (MAXE) values less than 0.6567 %, 0.6024 % and 2.2580 %, respectively. The results demonstrate the model's strong applicability for SOC estimation in both LIBs and SIBs, highlighting its potential for future transportation.and energy storage applications.