Enhanced bi-directional temporal convolutional gated recurrent hybrid neural network for state of charge estimation of power lithium-ion batteries

Abstract

With the rapid development of applications such as renewable energy and electric vehicles, accurate estimation of the state of charge (SOC) of lithium-ion batteries has become a key technology for improving system safety, prolonging battery life, and optimizing energy management. In order to cope with the complexity caused by the lack of battery modeling accuracy and varying environmental conditions, this paper proposes a data-driven hybrid neural network-based model. Specifically, a bidirectional time convolution network (BiTCN) is used to extract long-term features such as current and voltage. Then, a bidirectional gated recurrent unit (BiGRU) is used to predict the state of charge (SOC) of electric vehicle lithium-ion batteries. In order to obtain the globally optimal hyperparameter settings, a natural heuristic optimization algorithm crown porcupine optimizer (CPO) is introduced. The effectiveness of the fusion method was verified under different working conditions and temperatures. The average MAE for the DST condition is 0.54%, the average RMSE is 0.71%, and the average R² is 99.91%. The average MAE, RMSE, and R² for the BBDST condition are 0.72%, 0.92%, and 99.86%, respectively. The prediction performance is significantly improved compared with traditional machine learning methods, which is important for online charge state estimation and health management of electric vehicles.