Physics-Guided Deep Learning for Battery Future Capacity and Remaining Useful Life Predictions

IEEE Transactions on Transportation Electrification, vol. 12, pp. 4846–4860

Abstract

Accurate predictions of future capacity and remaining useful life (RUL) are critical for ensuring battery safety and prolonging lifespan. However, existing studies struggle to achieve high-accuracy long-term predictions from early cycles of battery operation. To address this limitation, this study proposes a physics-informed deep learning framework called physics-controlled differential equation (CDE) for forecasting battery capacity and RUL. The proposed approach integrates a Query-Key-Value neural CDE structure to extract attention-weighted features, effectively capturing essential battery dynamics. Furthermore, a long short-term memory-embedded neural CDE models differential relationships over time, while an empirical model periodically updates physical constraints to refine degradation trends. Experimental validation across multiple datasets demonstrates that the proposed method outperforms existing approaches by accurately reconstructing the entire capacity degradation curve from only a small number of early-cycle measurements, without requiring full-lifecycle capacity labels, while also providing reliable RUL predictions from early cycles. Even when forecasting capacity degradation over the full lifecycle on various datasets, the prediction error remains below 0.04, confirming the method’s robustness and precision.

Authors 4

1. Tianjing Wang

   City University of Hong Kong

   Affiliation as printed

   Department of Electrical Engineering, City University of Hong Kong, Kowloon Tong, Hong Kong

2. Weihan Li Aachen Institute for Power Electronics and Electrical Drives (ISEA)

   RWTH Aachen University

   Affiliation as printed

   Institute for Power Electronics and Electrical Drives, RWTH Aachen University, Aachen, Germany

3. Yan Hui Xu

   Nanyang Technological University

   Affiliation as printed

   School of Electrical and Electronic Engineering, Nanyang Technological University, Jurong West, Singapore

4. Zhao Yang Dong

   City University of Hong Kong

   Affiliation as printed

   Department of Electrical Engineering, City University of Hong Kong, Kowloon Tong, Hong Kong