Comparison of the performace of different Neural Networks Algorithm Functions in Simulation of Seasonal precipitation case study: Selected stations of khuzestan province

The precipitation is one of the main components of hydrology cycle. This complex phenomenon relates to several climatological factors. Over the last decades, Artificial Neural Networks (ANNs) have shown a considerable ability for modeling complex and nonlinear systems. The monthly rainfall data of three meteorological stations in Khuzestan province for 48 years, from 1961 to 2008 were used. Then, using these values as target outputs, various networks with different structures were defined and train. At last, the capability of the network for estimating precipitation was studied using a part of data which were not used for training the network. In present work, RBF and MLP networks with some changes in number of neurons and number of middle layers and MOM, LM, CG training algorithms were used to predict the seasonal rainfall. The results showed that for Ahvaz station, RBF network with 6-4-1 topology and LM learning have the highest correlation coefficient value, 0.96 and MSE had the lowest one, 0.044. For Abadan station, RBF network with 6-6-7-1 topology and LM learning had the highest value, both with 0.92 and MSE had the lowest one, 0.062. For Dezful station, MLP network with 6-3-4-1 topology and LM learning, both with 0.94, had the highest and MSE had the lowest value, 0.034.