Hybrid Electrochemical Modeling with Recurrent Neural Networks for Li-ion Batteries

This paper examines a hybrid battery system modeling framework, where data-oriented recurrent neural network (RNN) and first-principle electrochemical battery model are combined. The data-driven RNN model captures unmodeled dynamics in the electrochemical model. We specifically study a simple RNN model called an Elman network, which has feedback loops in the hidden layer. We analyze and prove convergence of the weight errors for a class of Elman networks and learning update laws. In simulation, we compare our proposed hybrid battery model with reduced electrochemical battery models. The results demonstrate that the proposed hybrid approach outperforms other reduced electrochemical battery models in most scenarios.