Prediction of effluent quality parameters of a wastewater treatment plant using a supervised committee fuzzy logic model

Abstract

Producing reclaimed water meeting water quality standards for agricultural and industrial demands is a viable option to the Tabriz area, East Azerbaijan, Iran, due to water scarcity. This study focuses on the Tabriz wastewater treatment plant (TWWTP) and investigates its treatability. This research proposes an ensemble of fuzzy logic (FL) models as surrogates for the TWWTP to avoid simulating complex physical, chemical, and biological treatment processes. Each FL model predicts water quality parameters of TWWTP using measured influent water quality data, such as biochemical oxygen demand (BOD), chemical oxygen demand (COD), pH, temperature, and total suspended solids. Instead of looking for the best FL forecasting model, this study introduces a supervised committee FL (SCFL) model as a predictive ensemble model for effluent water quality. The SCFL model uses an artificial neural network (ANN) to combine forecasted water quality results from individual FL models. Three FL models of Takagi-Sugeno, Mamdani, and Larsen are the surrogates for TWWTP. Compared to the historical data on effluent water quality, the individual FL models have a mean absolute percentage error (MAPE) for BOD, COD and TSS in the testing step ranging between 10% and 13%. Using the CFL model, the MAPE in the testing step reduces to 5%–7%. The SCFL model further reduces MAPE to 4% in the testing step. The SCFL model is an alternative to predict the quality of effluent water parameters by performing better than the individual FL and CFL models for the Wastewater Treatment Plant.