Based on the study of radial basis function (RBF) neural network training algorithm, a new RBF neural network training algorithm was introduced by combining genetic algorithm of chromosomes with changeable length and least-square method. It is able to determine the structure and parameters of network. The new training algorithm was used to model heat loading forecasting for co-generation power plants, compares with BP neural network and RBF neural network based on a training algorithm of automatic increase in hidden nodes. Simulation results show that the proposed method is valid.

Radial basis function neural network,   ,  Genetic algorithm,    , Least-square method,    ,  BP neural network,    , Heat loading