An online intelligent management method for wastewater treatment supported by coupling data-driven and mechanism models

Abstract

The management of traditional wastewater treatment plants(WWTPs) mostly rely on human experience, resulting in high cost and low efficiency under conservative operation mode. This study proposed an online intelligent management method based on plan library to improve the wastewater treatment performance. The operation plans in plan library were obtained based on the mechanism model (Activated Sludge Models) of a pilot Anaerobic-Anoxic-Oxic (A2/O) process, and each plan represents the optimal operation strategy under a specific influent condition. Then the plan library was input to train a data driven model, i.e. a Multi-Layer Perceptron (MLP) regression model. The trained MLP model can be used to generate more detailed online operation strategies for wastewater treatment process under different influent conditions. After a 124 days continuous operation of the A2/O process, the results shown, the proposed method can generate online operation strategies according to influent conditions, and the main effluent indicators meet the discharge standards continuously and stably. Meanwhile, compared with the manual operation mode, the aeration energy consumption was saved about 49.4 % by using the proposed method. The mechanism model and data-driven model were combined in this study, and it has scientific value and engineering significance for intelligent management of WWTPs.