Abstract: In order to solve the problem that pure electric vehicles are susceptible to current fluctuations and non-linear conditions during SOC prediction, a method for dynamic prediction of lithium battery SOC is proposed. Firstly, the parameter combination of the particle swarm clustering algorithm is optimized and combined with the preferred results to improve the radial basis function (RBF) neural network. Then, by analyzing the characteristics of the battery under different working conditions, the battery is divided into charging and static. Set and discharge three states. Different strategies are used to predict the SOC for the working state of the battery. In the battery discharge phase, the improved PSO-RBF algorithm is used to dynamically predict the SOC. in the battery standing and charging state, the two-point look-up table method is used to make the open circuit voltage curve considering the temperature drift and the current node abrupt curve during charging into two dimensions. Array table, the value of SOC is corrected by using the created two-dimensional array table. Thereby reducing system response time while improving accuracy. The experimental results show that the maximum error of the prediction correction model is about 1.9%, which verifies the effectiveness of the method.

Key words: SOC, particle swarm optimization, radial basis neural network, staged method