Mechanism-data-based hybrid modeling for denitrification in urban wastewater treatment

In order to overcome the shortcomings of the low accuracy of the mechanism model and poor physical interpretability of the data model, a mechanism-data-based hybrid model is designed for the denitrification process of urban wastewater treatment. To improve the applicability of the model, a SASM1 (Simplified Activated Sludge Model No.1) is set up by reasonably simplifying the denitrification process of urban wastewater treatment. To compensate the modeling errors of SASM1, a data-driven model based T-S fuzzy neural network is established by deeply excavating the characteristic relationships among the wastewater treatment data. A parallel structure that can take both advantages of the mechanism model and data model is designed. In order to show the effectiveness of the designed hybrid model, some comparisons between the SASM1 and former model are presented, where four water effluent indicators are considered. The result shows that the average prediction accuracy of effluent BOD has increased from 79.98% to 97.82%, and the average prediction accuracy of COD has increased from 31.92% to 98.68%, and the average prediction accuracy of other water effluent indicators has also been significantly improved.