Amirkabir University of TechnologyAmirkabir Journal of Civil Engineering2588-297X53820211023Application of Optimized Neuro-Fuzzy Models for Estimation of Water Quality Index in Karun RiverApplication of Optimized Neuro-Fuzzy Models for Estimation of Water Quality Index in Karun River34533466394210.22060/ceej.2020.17973.6725FAMohammadNajafzadehDepartment of Water Engineering, Faculty of Civil and Survey Engineering, Graduate University of Advanced Technology, Kerman0000-0002-4100-9699MuhammadLottfi-DashbalaghWater Engineering Department, Faculty of Civil and Surveying EngineeringJournal Article20200221Management of water quality is inextricably bound up with making good management decisions and this typical management is at the mercy of predicting the water quality index (<em>WQI</em>). The use of board range of artificial intelligence models for analyzing surface water quality is one of the most efficient techniques to predict water quality parameters and <em>WQI</em>. In the current research, at the first, datasets accumulated from nine hydrometry stations, located in Karun River, were included those of 13 water quality parameters (i.e., dissolved oxygen, chemical oxygen demand, biochemical oxygen demand, electrical conductivity, nitrate, nitrite, phosphate, turbidity, pH, calcium, magnesium, sodium, and water temperature) which was used to estimate <em>WQI</em>. So, to obtain an optimal selection of ANFIS model-feeding-input variables, gamma test (GT), forward selection (FS), and principal component analysis (PCA) evaluations were applied. Ultimately, constant coefficients of membership function used in the ANFIS model were computed by using evolutionary techniques including a genetic algorithm (GA), ant colony optimization (ACO), and particle swarm optimization (PSO) for training the structure of the ANFIS model. Results of statistical assessments indicated that the GT-ANFIS-PSO model with a correlation coefficient of 0.952, mean absolute error of 1.68, and root mean square error of 3.05 had a satisfying performance for prediction of <em>WQI</em> compared with other optimized ANFIS models. Moreover, values of <em>WQI</em> ranged from 30 to 58.4 which were indicative of being relatively poor to the good water quality of Karun River. Management of water quality is inextricably bound up with making good management decisions and this typical management is at the mercy of predicting the water quality index (<em>WQI</em>). The use of board range of artificial intelligence models for analyzing surface water quality is one of the most efficient techniques to predict water quality parameters and <em>WQI</em>. In the current research, at the first, datasets accumulated from nine hydrometry stations, located in Karun River, were included those of 13 water quality parameters (i.e., dissolved oxygen, chemical oxygen demand, biochemical oxygen demand, electrical conductivity, nitrate, nitrite, phosphate, turbidity, pH, calcium, magnesium, sodium, and water temperature) which was used to estimate <em>WQI</em>. So, to obtain an optimal selection of ANFIS model-feeding-input variables, gamma test (GT), forward selection (FS), and principal component analysis (PCA) evaluations were applied. Ultimately, constant coefficients of membership function used in the ANFIS model were computed by using evolutionary techniques including a genetic algorithm (GA), ant colony optimization (ACO), and particle swarm optimization (PSO) for training the structure of the ANFIS model. Results of statistical assessments indicated that the GT-ANFIS-PSO model with a correlation coefficient of 0.952, mean absolute error of 1.68, and root mean square error of 3.05 had a satisfying performance for prediction of <em>WQI</em> compared with other optimized ANFIS models. Moreover, values of <em>WQI</em> ranged from 30 to 58.4 which were indicative of being relatively poor to the good water quality of Karun River. https://ceej.aut.ac.ir/article_3942_b4d7eaba464bdc6953a4e9b882fcd1bc.pdf