MODELS FOR ACTIVATED SLUDGE PROCESS IN WASTEWATER TREATMENT PLANTS: DEVELOPMENT AND VALIDATION

The Activated Sludge Process (ASP) is the core of the biological treatment in a Wastewater Treatment Plant (WWTP). The Dissolved Oxygen (DO) concentration in the aeration tank is the key variable: it should be high enough to allow the degradation of organic matter and the conversion of ammonium into nitrate; at the same time it should not be excessively high, because this increases the energy consumption and lowers the efficiency of denitrification. The aim of the thesis is to identify a control-oriented model able to predict the DO level. The current plants are typically controlled by PID, manually tuned by operators. Therefore, the purpose of this dissertation is to provide an useful tool for designing and testing new types of controllers, such as MPC. These new controllers could improve the WWTP performances in terms of quality of the effluent, but also in terms of energy consumption, since the aeration is the most expensive process in a WWTP. Moreover, an accurate model of the aeration tank could be used as digital-twin to detect anomalies in real-time. Since white-box models of the Activated Sludge Model (ASM) family are difficult to estimate and it is difficult to apply them for control purposes, a data-driven approach based on NARX models is used. Four models are identified: two fully black-box NARX models (Model A and Model A2) and two NARX models whose regressors are derived from ASM1 processes (Model B and Model B2). All models perform well in a one-step prediction. The ones identified with a grey-box approach increase their performances on simulation. This shows that adding knowledge improves the prediction ability of the model. As an application of the developed models, a linear ARX model is used to design a MPC and it is evaluated in a closed loop simulation with the NARX Model A. In a two-day simulation, the designed controller keeps the DO concentration in the desired band in almost all of the time. This controller obtains better results in terms of tracking the DO than the current PID controller available in the plant.