SOC prediction method based on battery pack aging and consistency deviation of thermoelectric characteristics

Abstract

In view of the influence of different cell state parameters on the estimation of power battery packs’ state of charge (SOC), based on the travel data of electric vehicles in Beijing, random forest is used to reduce dimensionality, and the aging and thermoelectric characteristic parameters which have high correlation are selected as the input features of long short-term memory (LSTM). Then, using grid search to optimize the LSTM structure. Finally construct a data-driven method for high robustness prediction of battery SOC. The results show that the maximum absolute error (MaxAE) of the proposed SOC prediction method is only 1.539% under different temperatures, battery aging degrees and operating conditions. Compared with the two SOC prediction methods of gate recurrent unit (GRU) and recurrent neural network (RNN) , the MaxAE is reduced by 71.8% and 26.1% respectively. The research results provide a method basis for improving the robustness of power battery SOC estimation.

Keywords

• Electric vehicle
• SOC prediction
• Data driven
• Machine learning
• Deep learning