A fuzzy neural network-based soft sensor for modeling nutrient removal mechanism in a full-scale wastewater treatment system

Abstract

Abstract The nonlinearity and complicated biological phenomena existing in wastewater treatment processes (WWTP) make the operation and modeling of WWTP quite difficult. In this study, a hybrid learning method combining genetic algorithm with adaptive neuro-fuzzy inference system (GA-ANFIS) was serviced to estimate effluent nutrient concentrations in a full-scale biological wastewater treatment plant. The GA-ANFIS possessing a more flexible hybrid learning ability was adopted to capture the nonlinear relationships between the influent and effluent concentrations of pollutants. Having the capabilities of global and parallel optimization, GA was used to optimize the structure parameters of the fuzzy membership functions of GA-ANFIS. The real data collected from Korean Daewoo nutrient removal wastewater treatment plant were used to demonstrate the prediction efficiency of the proposed soft sensor with the aid of three performance indices of root mean square error, mean absolute percentage error, and squared correlation coefficient. The results indicate that the hybrid GA-ANFIS soft sensors outperform ANFIS-based soft sensors in terms of effluent prediction accuracy.