Multi-State Online Estimation of Lithium-Ion Batteries Based on Multi-Task Learning

Deep learning-based state estimation of lithium batteries is widely used in battery management system (BMS) design. However, due to the limitation on-board computing resources, multiple single-state models are more difficult deploy practice. Therefore, this paper proposes a multi-task learning network (MTL) combining multi-layer feature extraction structure with separated expert layers for joint charge (SOC) and energy (SOE) Li-ion batteries. MTL uses extract features, separating task sharing from task-specific parameters. The underlying LSTM initially extracts time-series features. layer, consisting shared experts, features specific different tasks tasks. information extracted by experts fused through gate structure. Tasks processed based on information. Multiple trained simultaneously improve performance learned knowledge each other. SOC SOE estimated Panasonic dataset, model tested generalization LG dataset. Mean Absolute Error (MAE) values two 1.01% 0.59%, Root Square (RMSE) 1.29% 0.77%, respectively. For tasks, MAE RMSE reduced 0.096% 0.087%, respectively, when compared single-task models. also achieves reductions up 0.818% 0.938% values, other 0.051% 0.078%, outperforms models, achieving 0.398% 0.578% In process simulating online prediction, consumes 4.93 ms, which less than combined time almost same as that results show effectiveness superiority method.