State-of-charge estimation for batteries based on temporal distribution characterization and matching transfer learning framework

Accurate estimation of the state of charge (SOC) and available capacity is critical for the efficient utilization and safety of batteries. This study addresses the temporal covariate shift that occurs during battery operation. A novel transfer learning framework based on temporal distribution characterization and matching is proposed. The temporal distribution characterization splits the source domain data into basic subsequences with low similarity, aiming to capture temporal information during battery charging and discharging. Temporal distribution matching aligns the model input data with subsequences of similar distribution, facilitating feature transfer and enhancing model generalization. This approach improves the model’s ability to handle distribution shifts in battery data under varying environmental conditions. Additionally, a Transformer model is integrated for SOC prediction. An online battery management system (BMS) for energy storage packs was developed to validate the model’s accuracy, incorporating diverse operating scenarios for real-time validation. Experimental results show that, compared to traditional algorithms, the proposed model achieves higher prediction accuracy and robust performance against various influencing factors and initial parameter errors. • TDC and TDM are proposed to address temporal covariate shift. • A novel transfer learning framework enhances model robustness. • The TDCM-TL-Transformer is proposed for SOC estimation. • The model achieves high prediction accuracy and robustness.