Battery SOH estimation based on thermodynamic parameters from an electrochemical fractional-order model and LSTM

Accurate state of health (SOH) estimation is vital for lithium-ion battery safety and performance. Since SOH cannot be directly measured, traditional methods rely on external features or models but often suffer from low interpretability and complex modeling. This study proposes a physics-inspired deep learning method for SOH estimation, which combines battery mechanism models with deep learning to effectively incorporate physical insights and improve predictive performance. Thermodynamic parameters from an electrochemical fractional-order model are fed into a long short-term memory (LSTM) network to map features to SOH, enabling high-precision estimation. Compared with four other machine learning methods, LSTM achieved the best performance, with an average root-mean-square error of 0.70% and a minimum error of 0.21% across eight validation batteries. The use of thermodynamic parameters improved estimation accuracy by 3.36 times compared to traditional features like incremental capacity curves and ohmic resistance. This approach integrates physical modeling with data-driven methods, enabling high-precision battery management.