نوع مقاله : مقاله کامل علمی پژوهشی
عنوان مقاله [English]
Infiltration rate is one of the most important soil physical parameters and is a basic input data in irrigation and drainage projects. Although, a number of theoretical or experimental based equations are presented to describe this phenomenon but the evaluation of some new sciences such as artificial neural networks, for prediction of the phenomenon can be investigated. Generally, the infiltration rate is a function of different soil factors, such as: organic materials; porosity; pH; EC; Na and Ca+Mg. Aburaihan campus belongs to theUniversityofTehran. It has two teaching and research centres, one of them is located in the Waramin lowlands of easternTehran(Ghezlagh farm). The farm has about 120 hectares of fertile lands. The present research plan aims to model the infiltration of the soils involving the Artificial Neural Networks (ANN) and the Statistics models. The performances of different types of neural networks, relevant functions and processing elements were examined using mean square error (MSE) as the criterion. The Multilayer perceptorns (MLPs, feed-forward network) with one hidden layer (three layers in total) including five neurons as neural network type and momentum as learning rule were the final option, which were chosen to built up the ANN model. Data from a previous study were used for this purpose. Also, a regression model involving the SPSS software has been used to predict the basic infiltration rate (Ibas). Results obtained from the artificial neural network and the regression models were compared in terms of correlation coefficient between measured and estimated values. The calculated correlation coefficients between the predicted and measured data were found to be 0.94 and 0.98 for the neural network and regression methods, respectively. Results indicated that the variable, Ibas, was predicted more efficiently by the regression model than the ANN model. However, from the encouraging results, it can be concluded that the use of a neural network model can be efficient for prediction of the basic infiltration rate.