Evaluation of input variables in adaptive-network-based fuzzy inference system modeling for an anaerobic wastewater treatment plant under unsteady state

A conceptual neural-fuzzy model based on adaptive-network-based fuzzy inference system (ANFIS) was proposed to estimate effluent chemical oxygen demand (COD) of a full-scale anaerobic wastewater treatment plant for a sugar factory operating at unsteady state. The fitness of simulated results was improved by adding two new input variables into the model; phase vectors of operational period and effluent COD values of last five days (history). In modeling studies, individual contribution of each input variable to the resulting model was evaluated. The addition of phase vectors and history of five days into the input variable matrix in ANFIS modeling for anaerobic wastewater treatment was applied for the first time in literature to increase the prediction power of the model. By this way, the correlation coefficient between estimated and measured values of output variable (COD) could be increased to the value of 0.8940, which is considered a good fit.