The Estimation of groundwater level changes using four different techniques of evolutionary neural network, case study of Dasht-e-Abbas plain, Ilam province

Abstract In dry and semi-arid areas, water is the most factor of limiter in agriculture. In these areas, due to the lack of surface flows, major pressures enter on groundwater. Groundwater resources in the studied area (Dashte-Abbas plain) also suffered a severe drop in surface water due to unplanned use. In this study, we compared four different models of evolutionary neural network, a multi-layered perceptron neural network with Genetic Algorithm (ANN-GA), a multilayered perceptron neural network with particle swarm optimization (ANN-PSO), a multilevel perceptron neural network with Imperialism competitive algorithm (ANN-ICA) and multi-layered perceptron neural network with ant colony optimization (ANN-ACOR) for estimating groundwater level according to groundwater inflow, effective penetration of rainfall, effective penetration of surface flow and flood, effective penetration of return water Agriculture, underground outflow, withdrawal from aquifer for agriculture, evaporation from groundwater level and past groundwater data Were used. groundwater level comparisons are the combination of inputs has been prepared using Auto-correlation analysis, partial Auto-correlation and cross-correlation for each model. Optimal models are obtained by changing the control parameters. The best results are obtained from the input models (GWLt-1, GWLt-2, Qint, Qpt-1, Qrt-1, Qit-1, Qoutt-1, Qwt-1, and Qet-1). The accuracy of the mean squared error in the test phase for ANN-PSO, ANN-ICA, ANN-ACOR models was 1.2208, 0.9456 and 1.7720, respectively, and for the ANN-GA model, it was 0.8739. The mean relative error of ANN-GA model is 3.6% and its determined coefficient is 0.9388. According to the results, the ANN-GA model showed better performance than the other three models for estimating groundwater level.