Evolutionary Artificial Neural Networks in Hydrological Forecasting

Abstract

The majority of artificial neural network (ANN) algorithms applied to different water resources problems are mainly the multi layer perceptrons with feedforward back propagation algorithm (FFBP). This method is sensitive to the initial random weight selection and is likely to be trapped by local minima. In this study another ANN algorithm, evolutionary artificial neural network (EANN) is employed in daily river flow forecasting. EANNs are the neural network structures into which the algorithms inspired of evolution are incorporated. In this implementation EANN has a multi-layer perceptron network structure but connection weights are adjusted by breeder genetic algorithm. The simulations showed that problems faced by back propagation algorithm were overcame and the EANN forecasts compared well with FFBP algorithm and the conventional stochastic model in terms of the selected performance criteria.