Context Neural Network for Temporal Correlation and Prediction

Abstract

This paper is focused on the application of Enhanced Neural Networks to the load demand forecasting. These nets can be considered as Context Networks since they are able to process the pattern set in order to obtain a valid context according to the input presented to the net, the context data are expressed in the weights of a neural network. Concerning the load demand forecasting, classical methods require two stages, first stage is a classification net to organize data, and second stage is a forecasting net to output desired response. With Context Neural Networks, only one stage is required. The net will perform a classifation and a forecasting process at the same time. Results of Context Networks against classical neural methods are compared, showing the improvement of proposed networks.