SocNet: A Physics-Guided Neural Network for Battery State-of-Charge Estimation Robust to Temperature Variations and Sensor Noises

Xilin Dai, Ruidi Zhou, Jinhao Zhang, Keyi He, Fanfan Lin, Hao Ma

Abstract: For lithium-ion batteries in electric vehicles (EVs), ambient temperature variations and sensor noises affect the accurate estimation of the state of charge (SOC). To achieve both temperature and noise robustness in SOC estimation with an efficient and interpretable approach, the SocNet is proposed in this article. As a physics-guided neural network (PGNN), the structure of SocNet is designed and optimized based on the physics knowledge, with the physics-guided feature extraction, SOC lookback, and linear summation layer. SocNet was trained and tested at different ambient temperatures, achieving a mean absolute error (MAE) of 0.40%. Furthermore, SocNet was tested with four different types of noise added to the input, and the MAE increased by only 0.28%. Compared with eight current methods, SocNet reduced the number of parameters by 68.06% and runtime by 49.12%, while improving temperature robustness and noise robustness by 43.66% and 76.87%.

External IDs:doi:10.1109/tte.2025.3573618