Assessment of GRNN model in comparison to ANN and RBF models for estimating confined aquifer parameters

Document Type : Research paper

Authors

1 M.Sc. Student, School of Engineering, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran

2 Professor, School of Engineering, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran

3 Assistant Professor, School of Engineering, Department of Civil and Environmental Engineering, Shiraz University, Shiraz, Iran

10.22034/hydro.2017.5475

Abstract

To achieve an appropriate management of groundwater resources, the accurate estimation of aquifer parameters is essential. In this study, several artificial intelligence models including, Artificial Neural Network (ANN), Generalized Regression Neural Network (GRNN) and Radial Basis Function (RBF) neural network have been developed to estimate the hydraulic parameters of a confined aquifer. One of the many reasons in using the artificial intelligence models to estimate the aquifer parameters, is their high flexibility, especially in non-linear problems. In order to implement these models, after gathering the pumping test data and reducing the dimensions of the data using the Principle Component Analysis (PCA), the artificial neural networks are trained and tested. Under the condition that the well function’s error as the output of the artificial intelligence models, lies within an acceptable limit, then the values of aquifer parameters can be determined. The models are applied to a pumping test data in a confined aquifer and the results are compared with those of the graphical Theis curve method. Several statistical errors considering the results of the proposed artificial intelligence models and the graphical Theis curve method are compared and the performance of the models is examined. As an example, the Mean Absolute Relative Error (MARE) in estimating aquifer parameters for ANN model and graphical Theis curve method are 0.5564 and 1.1320 percent, respectively. To this regard, the GRNN is more precise and has much less computational time than the others and may be selected as a superior model in the estimation of confined aquifer parameters.

Keywords