Abstract: To solve the problems of the large error, poor adaptability and complex modeling in predicting the remaining mileage of battery electric vehicles, firstly, the combined parameters for the clustering algorithm of particle swarm optimization (PSO) are selected. Then, the clustering for the number of hidden layer nodes q, the center vector ci and standardization constant δi of Gauss function in hidden nodes is conducted with PSO, the best combined parameters with least error and best clustering results are determined and timely updated according to different conditions for achieving the dynamic adaptability of prediction results. The modified PSORBF prediction model is established with terminal voltage, current, temperature and load as inputs and remaining mileage as output, the dynamic adaptive standard battery capacity and standard driving mileage in the fix and common environment of vehicle operation are defined, and the mathematical model for nonessential energy consumption is set up to correct the modified PSORBF prediction model. Finally, the results of comparison between test values and predicted values by different algorithms for EV1 battery electric vehicle show that the maximum relative error with RBF, PSORBF and modified PSORBF algorithms is 99%,62% and 38% respectively, demonstrating the significant enhancement in the accuracy of remaining mileage prediction by modified PSORBF, compared with other algorithms.

Key words: traction battery, BMS, remaining mileage, RBF, PSO