Neural network prediction of energy demand and supply in China

Traditional methods used to predict energy production and consumption analyse the relationship between cause and effect; such models thus tend to be complicated and not particularly accurate. Artificial neural network (ANN) methods are extensively used for prediction in non-linear systems; however, the back-propagation (BP) neural network (NN) method is not ideal for convergence and local minimum problems. The radial basis function (RBF) NN is capable of fast calculation and extrapolation, has a strong non-linear reflection function and is very adaptive. The objective of this paper is to apply a RBF NN to predict China's energy supply and demand. The core of the method is to process historic energy data as a static data series in order to simplify the model. All energy outputs depend only on historic data and the prediction requires a small calculation time. The results indicate that dynamic energy prediction based on a RBF NN shows good correlation with actual data and the method appears to be more appropriate for dynamic energy prediction than other models.