An Electrochemical Parameter Identification Method for Power Batteries Based on an Improved Physics-Informed Neural Network

Accurate identification of internal electrochemical parameters is essential for state estimation, model-based prediction, and safety warning of lithium-ion batteries. Conventional teardown-based methods are destructive and costly, while data-driven metaheuristic algorithms often suffer from low efficiency and poor generalization. To address these issues, this paper proposes a robust parameter identification method based on a Sequence-to-Sequence Physics-Informed Neural Network (SS-PINN). The governing equations of a reduced-order pseudo-two-dimensional model are embedded as physical constraints within a recurrent neural network. A hybrid loss function integrates voltage measurements with internal electrochemical mechanisms, enabling non-destructive and accurate identification of key geometric, transport, and kinetic parameters. Validation using real-vehicle dynamic operating data shows that the SS-PINN method achieves higher accuracy and faster convergence than the particle swarm optimization algorithm. The maximum voltage simulation error remains below 13 mV. Moreover, voltage curves generated with identified parameters align closely with those from direct teardown measurements, confirming experimental-level accuracy without battery damage. This study offers a promising solution for real-time state perception and adaptive parameter updating in battery management systems.