Short term electric load forecast using artificial neural networks

The application of neural networks (NN) in the area of power system engineering is increasing rapidly. Several researchers have suggested that short-term load forecasting (STLF) is a suitable area for the implementation of the NN approach. This paper presents work to further support two previous reports in which the authors described a method developed using NN and proposed a novel approach in applying NN to the problem of STLF. The proposed approaches to STLF have been developed for the PC environment as the primary target. With the increase in power and performance of networked PCs, such a platform is capable of supporting the large scale databases, high speed communications and processing power, required for the STFL process in the electricity supply industry (ESI). This improved method uses a NN architecture which consists of an ensemble of hidden layers, connected separately to a common input and output layer. The previously reported approaches are improved upon and enhanced to provide 3 hour ahead half hourly forecast together with a sliding window of validated system demand data applied directly to the input layer of the NN. The network includes a data filter that has been developed to remove the generalised noise from the input data set, which is used as an additional input for the NN. Results from annual demand curves with the corresponding forecasts for comparative purposes, which show that the NN approach achieves a high degree of accuracy, comparable with values reported in the literature for STLF, are presented