Short-Term Load Forecasting Based on RBF Neural Network

Abstract

In order to fully explore and analyze the inherent law of power load data and improve the prediction accuracy of residents' daily electricity consumption, because of analysis and comparison of multiple forecasting methods, the radial basis function (RBF) neural network algorithm with the introduction of clustering idea is used in this paper. The historical load and daily maximum temperature are chosen to cluster by k-means algorithm, according to change rule of the example data and the correlation between load data and other attributes, such as temperature, weather, and holiday. Then, the centres of each cluster in the sample is taken as the centres of the hidden layer of the RBF neural network to realize the training of sample data and forecast short-term power load. The experiment shows this method has higher accuracy in short-term load forecasting compared with time series method and BP neural network.