A physics-informed neural network approach to parameter estimation of lithium-ion battery electrochemical model

Abstract

The electrochemical models (EMs) have gained increasing attention for their ability to reflect the internal states of the lithium-ion battery, yet the parameter identification is still an essential and tricky task. This paper proposes a novel physical-informed neural network (PINN) framework for the parameter identification of the EM, where three neural networks are operated in parallel to search the suitable microscopic parameters. Compared with existing neural network-based methods, PINN can utilize the prior knowledge of battery such as the physical laws that govern the system. To reduce the dimension of searching space, a parameter categorization approach is designed where parameters can be identified based on their properties under different discharge conditions. Afterwards, the PINN is integrated with EM to achieve the model prediction and parameter identification. The validation results demonstrate the good performance of the proposed method in both parameter identification and model prediction under various scenarios.