Calibration of electrochemical, thermal, and aging parameters for a physics-based lithium-ion battery model assisted by a data driven approach

As the demand for lithium-ion batteries (LIBs) continues to rise, the need for physics-based models capable of accurately assessing internal battery states has become increasingly critical. A major challenge in such modeling lies in the precise calibration of parameters governing electrochemical, thermal, and aging behaviors. This study introduces a genetic algorithm-based optimization strategy to estimate 14 sensitive parameters used in the enhanced single particle model (ESPM), a widely adopted physics-based model for LIBs. The approach leverages a data driven surrogate model built on an artificial neural network (ANN), designed to replicate complex interactions typically captured by physics-based simulations. Compared to direct ESPM computations, the surrogate model achieved a 420 times speed increase under constant current scenarios and an extraordinary 23,970 times improvement during cycle testing. In terms of mean absolute error, the surrogate model demonstrated high precision relative to the ESPM, with deviations constrained to 0.425 mV for voltage, 0.01 °C for temperature, and 0.006 % for state of health (SOH). Parameter optimization targeted voltage, temperature, and SOH across diverse operating conditions, achieving root mean square error values consistently within 30 mV, 0.5 °C, and 0.1 %, respectively.