Optimized Training Options of a Deep Learning Model for Lithium-Ion Battery State of Charge Estimation

Lithium-ion batteries are currently considered as an ideal power source for electrified transportation and energy storage systems. Accurately estimating the state of charge (SOC) of batteries is crucial for the safety and reliability of battery operations. Due to the non-linear and time-varying relationship between the measured battery voltages and their corresponding SOCs, it urgently requires an intelligent approach to accurately estimate the SOC. This paper optimizes the training options of a deep learning model, the gate recurrent unit model, for battery SOC estimation method. The establishment steps of a deep learning model is firstly presented, followed by presenting the SOC estimation results based on the initial model. Different training options of the deep learning model, including the epoch, learning rate parameters, and activation function, etc., are then systematically analyzed. Through the comparative analysis of different training options, optimized parameters are applied in the model to improve the accuracy of battery SOC estimation. Finally, the optimization rate is proposed to quantitively evaluate the improvement of the optimized model. Verification results indicate that the optimized model can effectively decrease the estimation errors compared with the initial model.