Modeling of Groundwater Fluctuations Based on Artificial Intelligence Methods (Case study: Zawah-Torbat Heidarieh plain)

Groundwater resources are one of the most important water sources in each country. That proper knowledge and basic exploitation in this field can play a principal role in the sustainable development of the social and economic activities of a region, especially in semiarid and dry areas. The prediction of groundwater level fluctuations for supplying management and exploit Akon of watering is essential the purpose of this research is to predict Zawah-Torbat Heidarieh groundwater level32 fluctuations with a range of about 2054 square kilometers is located in the north of the desert pans on desert flats in, south of Mashhad. In order to training of the model, information from 18 piezometers extracted by the researchers of this study, which had a staggered surface alignment level with a time series of 20 years (1375-1395), was used. Each piezometer is registered on a monthly basis with a delay of t0-1 (last month), and in each piezometer, seven parameters form the system inputs. For process modeling, multi-layer perceptron neural networks with error propagation algorithm and LVQ network are used. The calculation error is calculated using the least squares method (MSE). The amount of groundwater level is also the only output of this neural network. The results of this study showed that the artificial neural network with the Gradient Descent, Gradient Descent With Momentum, Levenberg Marquardt algorithms was able to predict groundwater levels in the monthly interval is (RMSE=0/0012) in the training phase and is (RMSE=0/021) in the testing  phase in the study area.