Improving neural networks prediction using fuzzy-genetic model

This paper proposes a new technique for improving artificial neural network (ANN) prediction using fuzzy-genetic model. The proposed method is applied to the prediction of daily natural gas consumption needed by gas utilities. A hybrid system consists of two stages, with the first stage containing two ANN predictors. Both predictors are a multilayer feed-forward network trained with back-propagation, but the first one is trained to predict daily natural gas consumption while the second one is trained to predict the change in the daily natural gas consumption from previous day. These two separate predictors are combined in the second stage using a fuzzy-genetic combiner. An adaptive scheme is used to change the weights of the ANN predictors throughout the prediction phase. The performance of the ANN predictors and the combination method is tested on real data from four different gas utilities for a period of several months. The results show that the proposed fuzzy-genetic combiner result in more accurate prediction.