A method for predicting nonlinear time series using RBF (Radial Base Function) neural network and its application

An innovative method revolving the use of RBF (radial base function) neural network based on a training algorithm of automatic increase in hidden nodes has been employed to predict nonlinear time series. The proposed method allows to successfully tackle the problem of selecting local minimal hidden node number and excessive fitting in BP networks. It has been applied to predict the thermal loads of a thermal power plant. The results of prediction indicate that very satisfactory results have been achieved in forecasting the thermal loads of power plants.