Genetic optimization of a self organizing fuzzy-neural network for load forecasting

In this paper a self-organizing fuzzy-neural network with a new learning mechanism and rule optimization using genetic algorithm (GA) is proposed for load forecasting. The number of rules in the inferencing layer is optimized using a genetic algorithm and an appropriate fitness function. We devise a learning algorithm for updating the connecting weights as well as the structure of the membership functions of the network. The proposed algorithm exploits the notion of error back propagation. The network weights are initialized with random weights instead of any preselected ones. The performance of the network is validated by extensive simulation results using practical data ranging over a period of two years. The optimized fuzzy neural network provides an accurate prediction of electrical load in a time frame varying from 24 to 168 hours ahead. The algorithm is adaptive and performs much better than the existing ANN techniques used for load forecasting.