Radial-based-function neural network based SOC estimation for electric vehicles

A generalized growing and pruning radial based function (RBF) neural network approach is used to estimate the battery's state of charge of the electric vehicles. According to the different input vector of network, several estimation models including standard model, recursion model, and ampere-hours model are developed for the estimation of the state of charge. The experiment shows that all models are able to effectively estimate the state of charge (SOC) at given working voltages, currents and surface temperature. And the performance of these estimators are compared in terms of their accuracy, training time, and network structure complexity. The results show that ampere-hours model is the best estimation method because of its higher accuracy, less training time, less complexity. Meanwhile, compared with the original generalized growing and pruning radial based function neural network training method, using decouple extended Kalman filter algorithm can save half of the training time without lowering precision.