Optimization and Prediction Changes of Groundwater Quality Parameters Using ANN+PSO and ANN+P-PSO Models (Case Study: Dezful Plain)

One of the main aims of water resource planners and managers is the estimation and prediction of groundwater quality parameters to make managerial decisions. In this regard, many models have been developed which proposed better managements in order to maintain water quality. Most of these models require input parameters which are hardly available or their measurements are time consuming and expensive. Among them, Artificial Neural Network (ANN) models inspired by human's brain are a better choice.

The present study stimulated the groundwater quality parameters of Dezful plain including Sodium Adsorption Ratio (SAR), Electrical Conductivity (EC), Total Dissolved Solids (TDS), using ANN+PSO and ANN+P-PSO models and in the end is comparing their results with measured data. The input data for TDS quality parameter consist of EC, SAR, pH, SO4, Ca, Mg and Na, for SAR including the TDS, pH, Na, Hco3 and quality parameter of EC contains So4, Ca, Mg, SAR and pH, gathered from 2011 to 2015.

The results indicated that the highest prediction accuracy of quality parameters of SAR, EC and TDS is related to the ANN+P-PSO model so that the MAE and RMSE statistics have the minimum and  has the maximum value for the model. The results showed that RMSE for PSO in predicting SAR, EC and TDS were 0.09, 0.045 (µs/cm) and 0.053 (mg/l) in testing period, respectively. These statistical criteria were 0.039, 0.031 (µs/cm) and 0.045 (mg/l) for P-PSO in this period, respectively.

The results showed that P-PSO had more accuracy compared to PSO. In addition, there were no significant differences between ANN and collecting values. So, it is recommended that ANN were applied to determine nitrate concentration in groundwater.