Abstract

Recently, Artificial Neural Network has been widely used in natural gas consumption forecasting in addition to classical time series analysis. In this study, RBF (Radial Basis Function) neural network which is a well known function approximation is used to predict natural gas consumption. In the case of having large dataset, how many neurons will be used in the hidden layer and how the centers that are belong to neurons located in this layer are determined are important issues. To overcome these problems, SOM artificial neural network and K-means clustering algorithm have been utilized in training RBF. The predictions are made by means of RBF that is trained with both methods and results are compared.