Lithium-ion battery state of charge prediction under multiple aging stages: an animated oat optimization algorithm-optimized temporal convolutional network-attention model study

Abstract

Accurate state of charge (SOC) estimation for lithium-ion batteries across their entire lifespan is crucial for battery management systems (BMS) but remains challenging due to capacity degradation. To enhance the robustness of SOC prediction under multiple aging stages, this study proposes a novel hybrid model named AOO-TCN-Attention. The model integrates a temporal convolutional network (TCN) to capture long-term temporal dependencies, an attention mechanism to adaptively focus on critical features, and the animated oat optimization (AOO) algorithm to globally optimize the hyperparameters of the TCN-Attention structure. The model was validated using a dataset from A123 18,650 cells aged over 3200 cycles under rigorous conditions. The results demonstrate that the AOO-TCN-Attention model achieves superior performance, with the root mean square error (RMSE) and mean absolute error (MAE) remaining below 2.25 % and 1.8 %, respectively, and the coefficient of determination (R²) exceeding 98.95 % at all tested aging cycles. In comparative tests, the proposed model significantly outperformed both long short-term memory (LSTM) and standard TCN models. Its exceptional generalization capability was further confirmed on a more demanding dataset, where it maintained robust performance (RMSE < 2.5 %, MAE < 2.0 %, R² $\approx$ 99 %) even at severe aging states. This study provides a highly accurate and reliable solution for SOC estimation throughout a battery's entire lifecycle.