Abstract: This article proposed a model improvement method based on the second-order RC equivalent circuit model,which was carried out under testing conditions of missing temperature sample and fewer discharge current sample. And a BP neural network was constructed to predict the terminal voltage error of the model in a wide range of temperature and current rates,in order to achieve dynamic compensation of the model. This method can avoid the loss of cycle life caused by repeated charging and discharging experiments of lithium-ion batteries. The simulation and experimental results indicate that the proposed small sample data expansion method is feasible;The battery model with an additional controllable voltage source has better adaptability at different temperatures and current rates,improving the accuracy of the model. Through model analysis,batteries are screened to form a cascade utilization energy storage system,and combined with battery models to achieve SOC prediction and heat warning. The model has practical engineering significance for the application of new energy generation battery energy storage systems.