TIME SERIES FORECAST WITH ELMAN NEURAL NETWORKS AND GENETIC ALGORITHMS

This chapter investigates into recursive neural networks and their application in time series forecast. As one of the most popular recurrent neural networks, an Elman neural network is studied in this chapter. It has been proven that the Elman network is able to approximate the trajectory of a given dynamic system for any fixed length of time. This ability is explored in the area of time series forecasting. The electricity market demand signal, as a typical time series, is studied in the chapter with Elman networks. In order to obtain the best available optimal weight allocation, a Genetic Algorithm (GA) is used to train the recurrent neural networks in the forecast model. The forecast simulation is carried out on electricity market load data series with Elman networks as well as GA trained Elman networks to compare their performance.