A radial basis function neural network based multi-objective optimization for simultaneously enhanced nitrogen and phosphorus removal in a full-scale integrated surface flow treatment wetland-pond system

Abstract

In this study, a radial basis function neural network (RBFNN) model was developed and implemented in a multi-objective optimization procedure to determine the optimal hydraulic loading rate (HLR), hydraulic retention time (HRT), and mass loading rates (MLR) for enhanced removal of nitrogen and phosphorus by an integrated surface flow treatment wetland-pond system treating drinking source water in Yancheng, China. Prior to modelling, the system's 6-year nitrogen and phosphorus removal efficiencies were found to trend downwards as effluent concentrations trended positively. Meanwhile, operating parameter interaction effects impacted final effluent quality. Thus, total nitrogen and total phosphorus removal were simulated by an RBFNN model with satisfactory R² of 0.99 and 0.98 respectively. Optimal average HLR, HRT and MLR for 80% simultaneous removal efficiencies were subsequently determined to be 0.10860 ± 0.03 md^-1, 30.43 ± 9.96 d and 306.416 ± 89.54 mgm^-2d^-1 respectively. The results highlight the feasibility of the RBFNN modelling based optimization procedure for treatment wetlands.

Keywords: Multi-objective optimization; Nutrient removal; Operational parameters; Radial basis function neural network; Treatment wetlands.