A variable-weight combination forecasting model based on GM(1,1) model and RBF neural network

Abstract

A variable-weight combination forecasting model using the least square method is built for solving, which is based on grey GM(1,1) model and RBF neural network. With actual consumption data, these three models can be used to predict the monthly social total electricity demand of a year for the particular area respectively. Through comparing the actual load value with the prediction results obtained by different models, predicted value, the actual value graphical trend and relative error of the prediction results obtained in the three models are analyzed. The feasibility of three load forecasting models, which are applicable to 'small samples' object is discussed. In MATLAB simulation, using actual load data to predict, it's borne out that the outcome of the variable weight combination forecasting is better than the gray prediction method and RBF neural network prediction method and it is suitable for the selected region of the actual situation in the text.