| Abstract:
One of the important issues concerning the rivers quality is prediction of Total Dissolved Solids (TDS) in water. In this study, the performance of Support Vector Machine (SVM) intelligent models with different kernel functions, Gene Expression Programming (GEP) and Bayesian Networks (BN) were investigated in estimating the TDS in Kashkan River water. For this purpose, the quality data of Poldokhtar Station located in Lorestan Province were utilized to predict the TDS in water during the statistical period of 1991-2016 which included the hydrogen carbonate, chloride, sulfate, magnesium, calcium, sodium, electrical conductivity, flow rate and pH. To validate the models the criteria such as the coefficient of determination (R2), Nash-Sutcliff coefficient, Root Mean Square Error (RMSE), and bias coefficient were applied. The obtained results showed that in all three models, the combinational structures have acceptable accuracy. Also, according to the assessment criteria, it was found that the SVM with Radial Basis Function (RBF) kernel has the highest accuracy of 0. 982, minimum root square error of 0. 032 mg/lit, minimum bias of 0. 001 and highest Nash-Sutcliffe coefficient value of 0. 963 with respect to other models.
|