A physics-constrained Informer-LSTM network for battery state-of-charge estimation
Yi Xie, Zhenqi Han, Jialu Zhang, Lizhuang Liu
Abstract
Lithium-ion batteries are widely used in energy storage systems due to their high energy density and long cycle life. As a key indicator of remaining battery capacity, the state-of-charge (SOC) is crucial for battery management system (BMS). However, SOC cannot be directly measured and is affected by various nonlinear factors, posing significant challenges for accurate estimation. To enhance estimation accuracy and generalization, this study proposes a deep learning model incorporating physical constraints. The model couples the Informer and long short-term memory (LSTM) architectures: the Informer extracts global temporal features from battery data, while the LSTM models key parameters of a second-order equivalent circuit, embedding them as prior knowledge into the SOC estimation process to ensure physical consistency. Additionally, a temperature-aware weighted loss function is introduced to improve the model’s robustness under complex operating conditions. Experimental results demonstrated that the proposed method outperformed existing approaches in both estimation accuracy and stability, showing strong potential for practical engineering applications.