Development of PSO-SVR and ANN models to evaluate the effect of cross-sectional shape on denitrifying bioreactors

ABSTRACT Nitrate measurement is a complex, costly, and time-consuming task. Predictive methods such as modeling offer a feasible solution to these challenges. Therefore, this study aimed to develop a hybrid simulation–optimization model to predict nitrate removed from different cross-sectional shapes of denitrifying bioreactors. The model employed particle swarm optimization (PSO) to optimize support vector regression (SVR) parameters, and the results were compared with those of artificial neural networks (ANNs). The results demonstrated that the PSO-SVR model performed better than ANN in predicting nitrate removal. Based on PSO-SVR, triangular cross-section provided better prediction when the salinity level was 10 ds/m with (R-square = 0.95, 1.00, and 0.97, and RMSE = 0.57, 0.09, and 0.23 for training, validation, and testing, respectively) compared with ANN modeling best case that was triangular cross-section with the salinity level of 1 ds/m (R2 = 0.90, 0.94, and 0.96 and RMSE = 0.71, 0.70, and 0.60 for training, validation, and testing, respectively). Overall, the PSO-SVR model demonstrated a high level of accuracy in predicting nitrate concentrations, making it a precious implement to optimize the performance of denitrifying bioreactors and confirming the importance of effective pollutant removal strategies in wastewater treatment.