Prediction the performance of multistage moving bed biological process using artificial neural network (ANN)

Abstract

Complexity, uncertainty, and high dynamic nature of nutrient removal through biological processes (BPs) makes it difficult to model and control these processes, forcing designers to rely on approximations, probabilities, and assumptions. To cope with this difficult task and perform an effective and well-controlled BP operation, an artificial neural network (ANN) algorithm was developed to simulate, model, and control a three-stage (anaerobic/anoxic and MBBR) enhanced nutrient removal biological process (ENR-BP) challenging real wastewater. The effect of surface area loading rate (SALR), organic matters (OMs), nutrients (N & P), feed flow rate (Qfeed), hydraulic retention time (HRT), and internal recycle flow (IRF) on the performance of the ENR-BP to fulfil rigorous discharge limitations were evaluated. Experimental data was used to develop the appropriate architecture for the AAN using iterative steps of training and testing. Significant removals of chemical oxygen demand (COD) (89.2 to 98.3%), NH4+ (88.5 to 98.9%), and total phosphorus (TP) (77.9 to 99.9%) were achieved at a total HRT of 13.3 h (HRTZ-1 = 3 h, HRTZ-2 = 6 h and HRTZ-3 = 5.3 h) and an IRF value of 1.75. The ENR-BP treatment mechanism relies on the use of OMs as a source of energy for phosphorus bio-uptake and the simultaneous nitrification and denitrification (SND) of nitrogen compounds. The removal efficiencies in the proposed ENR-BP were four fold higher than the suspended growth process and in the same order of magnitude of 5-stage Bardenpho-MBBR. The developed ANN-based model provides an efficient and robust tool for predicting and forecasting the performance of the ENR-BP.