Modelling of Phosphorusremoval in Integrated Constructed Wetlands Using Adaptive Neuro-Fuzzy Inference System

Mawuli Dzakpasu, Miklas Scholz, Valerie McCarthy, Siobhán Jordan



Adaptive neuro-fuzzy inference system (ANFIS) models were developed to elucidate the removal of molybdate reactive phosphate (MRP) and to predict effluent concentrations in integrated constructed wetlands (ICW). The ANFIS models were developed and validated with a four-year data set from a full-scale ICW treating domestic wastewater. The models highlighted the importance of physicochemical parameters in the removal of MRP in ICW systems. High water temperature, dissolved oxygen, and electrical conductivity; and low oxidation-reduction potential were associated with high rate MRP removal. Findings indicate that ANFIS could predict the effluent MRP variation quite strongly. The simulated effluent MRP concentrations well fit measured concentrations. Effluent MRP were predicted to a reasonable accuracy (MASE = 0.12) by using input variables which can be easily monitored in real time as sole predictor variables. The validated model could be a useful tool for the rapid estimation of MRP removal in ICW systems. The rapid prediction with ANFIS provides an inexpensive alternative to laborious laboratory analyses and also serves as a management tool for day-to-day process control of ICW systems.