Artificial Neural Network (ANN) based Model Predictive Control (MPC) for an Experimental Dissolved Air Flotation Process

Abstract

Turbidity is one of many factors that affect the quality of drinkable water. While conventional filters may improve the condition of water, they require a lot of wash fluid for the process, making them less economical. For that reason, dissolved air flotation is a preceding step in reducing water turbidity. It does so by injecting air bubbles generated from a supersaturated solution of pressurized air, which arrives and contacts the contaminating solid, causing a decrease in the apparent density of these particles by forming a particle/micro-bubble agglomerate. The equipment performance can be evaluated considering environmental standards. SABESP recommends that the turbidity for drinking water treatment should not exceed 5 NTU, so it is plausible to adopt a setpoint tracking strategy considering stability, deviation from setpoint, and control effort as metrics. However, it is difficult to determine the relations between the variables involved in the process. Many mathematical models rely on strong approximations that may differ from the real process dynamics. With that, it is possible to identify the system and develop a control strategy based on artificial neural networks and their ability to capture nonlinear relationships between the variables and model predictive control to systematically ensure that there are no violations of system constraints and provide stability under a nonlinear system.