A Physics-Informed Integrated Modeling Method for Lithium-ion Batteries.

Battery models play a crucial role in battery management systems, describing the inner workings of batteries. However, offline models cannot adapt to the battery parameter degradation. To address this issue, a physics-informed integrated battery model combines equivalent electrical circuits and a neural network regression model is proposed. Electrical circuits capture essential battery electricity behaviors while a regression model with the recurrent neural network is employed to simulate the nonlinear open circuit voltage (OCV). The regression model is trained using unlabeled learning and the parameters of the electrical circuit are identified online. Finally, the proposed battery model is validated with a cubature Kalman filter. Extensive experiments confirmed that the proposed modeling method outperforms other methods with Less than 1.66% RMSE in SOC estimation.