Modeling groundwater fluctuations by three different evolutionary neural network techniques using hydroclimatic data

The accuracies of three different evolutionary artificial neural network ( ANN) approaches, ANN with genetic algorithm (ANN-GA), ANN with particle swarm optimization (ANN-PSO) and ANN with imperialist competitive algorithm (ANN-ICA), were compared in estimating groundwater levels (GWL) based on precipitation, evaporation and previous GWL data. The input combinations determined using auto-, partial auto- and cross-correlation analyses and tried for each model are: (i) GWL(t-1) and GWL(t-2); (ii) GWL(t-1), GWL(t- 2) and P-t; (iii) GWL(t- 1), GWL(t- 2) and E-t; (iv) GWL(t- 1), GWL(t- 2), P-t and E-t; (v) GWL(t- 1), GWL(t- 2) and Pt-1 where GWL(t), P-t and Et indicate the GWL, precipitation and evaporation at time t, individually. The optimal ANNGA,ANN-PSO and ANN-ICA models were obtained by trying various control parameters. The best accuracies of the ANN-GA, ANN-PSO and ANN-ICA models were obtained from input combination (i). The mean square error accuracies of the ANN-GA and ANN-ICA models were increased by 165 and 124% using ANN-PSO model. The results indicated that the ANN-PSO model performed better than the other models in modeling monthly groundwater levels.