Estimating Hydraulic Performance of Pressurized Irrigation System Using Artificial Neural Networks and Nonparametric Regression

In this research, replacement of hydraulic models with statistical models and artificial neural networks were studied in order to estimate the criteria of pressurized irrigation systems hydraulic performance. The Coefficient of Uniformity of Christiansen (CU) was accepted as a hydraulic performance index. Using an automated algorithm, the values of this index were calculated for different combinations of inlet pressure, number and spacing of outlets, pipe roughness coefficient, inside diameter, slope, outlets nominal outflow and pressure and the exponent of the formula of outlet outflows (x) (4320 different combinations). Two different architecture of artificial neural networks were studied including a multi-layer perceptron (MLP) model and a generalize regression model (GRNN). Again, K-nearest neighbor (KNN) algorithm, as a nonparametric regression model was analyzed too. The results showed that MLP model could estimate the CU values of pressurized irrigation system laterals very closely (2-3% error) using its hydraulic and physical characteristics. The performance of GRNN model was also acceptable, especially related to the whole data set. But, the KNN algorithm was unable to predict standard deviation of CU values, although it was capable in estimating the mean value. The deviations of the KNN algorithm were the largest among all the models. The lowest values of error indices of the KNN algorithm was related to the K values of 10 and 15. The results of this study revealed the possibility of simplification of sophisticated hydraulic models by replacing the whole or some parts of these models with simpler statistical models and artificial neural networks. This is very interesting because of the complexity of hydraulic models, especially in optimization processes of irrigation systems.