Comparison of Artificial Neural Network (ANN) and Multi Variable Regression Analysis (MRA) Models to Predict Ground Water Quality Changes (Case Study: Kashan Aquifer

Document Type : Research Paper

Authors

Abstract

The adjacency of Kashan aquifer to the saltwater front of the Salt Lake has caused a hydraulic gradient, resulting in the advancement of saltwater into the aquifer. Owing to the current situation, qualitative simulation of groundwater of Kashan plain has been implemented with Artificial Neural Network and Multi Variable Regression models in this study. For this purpose, prior to the model implementation, first we attempted to determine the dominant type of water. Results showed that the sodium chloride was the dominant type of water. Therefore, in addition to the water table fluctuations and precipitation amount, the chloride concentration in the previous year was considered as the model's input, while the output was the chloride concentration in this year. The results indicated that the MLP produced more accurate results than the RBF and MLR models, so that, the corresponding adjusted R2 values for these models were 0.97, 0.89 and 0.34, respectively. The outcomes revealed that the linear hyperbolic tangent activation function and Momentum algorithm produced better results than the other applied algorithms and functions. The resulted outcome of sensitivity analysis showed that concentration of chloride in the previous year and water table fluctuations had the most effect on the chloride concentration simulation.
 
 

Keywords

Water and Soil Science
Volume 25, Issue 2 - Serial Number 2
October 2015
Pages 207-220

Files

Share

How to cite

Statistics