Prediction of water quality parameter in Jajrood River basin: Application of multi layer perceptron (MLP) perceptron and radial basis function networks of artificial neural networks (ANNs)

Abstract

River water quality is a significant concern in many countries, considering agricultural and drinking consumptions. Therefore, prediction of salinity index, as the main water quality condition is a necessary tool for water resources planning and management. This paper describes the application of artificial neural networks (ANNs) models for computing the total dissolved solids (TDS) level in Jajrood River (Iran). Two ANN networks, multi-layer perceptron (MLP) and radial basis function (RBF), were identified, validated and tested for the computation of TDS concentrations. Both networks employed five input water quality variables measured in river water over a period of 40 years. The performance of the ANN models was checked through the coefficient of determination (R2) and root mean square error (RMSE). Jajrood River is one of the most important rivers which is located adjacent to Tehran city and supplies drinking water for people who live in this mega-city and recreational uses. Tehran is the most populous city and largest industrial pole in Iran, which caused the river, to be exposed to various pollutants. Matlab 2007 was selected for modeling goals in this research. Results show that MLP and RBF modeling as two methods of ANN are able to simulate water quality variables of Jajrood River with more than 90% accuracy. After modeling in MLP and RBF formatting and comparing simulation results (output) show that, the RBF result (R2 of validation is 0.9362) are more closely to reality than the MLP (R2 of validation is 0.8968). In other words, because of large number of input data, the RBF modeling performance has a better prediction than MLP modeling. Key words: Water quality variables, artificial neural network (ANN), multi-layer perceptron (MLP), radial basis function (RBF).