Modeling Groundwater Quality Parameters Using Hybrid Neuro-Fuzzy Methods

In this study, the application of four evolutionary algorithms, continuous genetic algorithm (CGA), particle swarm optimization (PSO), ant colony optimization for continuous domains (ACO(R)), and differential evolution (DE) were considered for training and optimization of adaptive neuro-fuzzy inference system (ANFIS) to model groundwater quality variables. At first, using correlation and sensitivity analysis, the best inputs were selected to estimate electrical conductivity (EC), sodium adsorption ratio (SAR) and total hardness (TH). After that, the quality variables were modeled by simple ANFIS and the ANFIS trained by evolutionary algorithms. Finally, the models' performances were evaluated using determination coefficient (R-2), root mean square error (RMSE), and mean absolute percentage error (MAPE) and sensitivity analysis. Results indicated that: 1) All the suggested algorithms improved the ANFIS performance in the modeling of EC and TH. Also, in SAR, CGA and PSO had a better performance than existing algorithms of ANFIS. 2) CGA with the most appropriate results, was the best algorithm in improving ANFIS performance for modeling the groundwater quality variables such that the amounts of R-2, RMSE, and MAPE were improved by 0.14, 35.4, and 0.59 for TH, by 0.13, 226 (mho Cm-1), 2.16 for EC, and by 0.15, 690, and 19.04 for SAR, respectively. 3) Sensitivity analysis showed that the results obtained by correlation analysis was dependable and could be used as a primary step in choosing the best input data for prediction of groundwater quality variables.