AN ARTIFICIAL NEURAL NETWORK BASED MODELS FOR SHORT TERM ELECTRIC LOAD FORECASTING

摘要

This paper presents a nonconventional, based on artificial neural networks (ANNs) approach to modeling and short-term power demand (load) forecasting. A feedforward multilayer perceptron (MLP) model as a predictor alone and in combination with the Kohonen neural network (Self-Organizing Map)is proposed. In order to train the MLP two basic approaches, for comparision, have been used namely standard learning procedure and accumulating (batch) one with a modified conjugate gradient optimization method as a nonconventional back-propagation learning algorithm.