Model predictive controller design with process constraints and implicit economic criteria gasoline blending

摘要

A method for designing a QDMC (Quadratic/Dynamic Matrix Control) controller is developed such that, at steady-state, it will maintain the process near the economic optimum, without explicitly calculating profit as part of its objective function. The method for designing the QDMC controller involves selecting the adjustable parameters of the QDMC controller such that the total profit is maximized for a given set of disturbances; this set of disturbances acts as a "training set" for the controller. The steady-state performance of an optimally-tuned QDMC controller was compared against both the optimum profit case and multi-loop PID control of a gasoline blending process with five manipulated and three controlled variables. The QDMC controller outperformed the multi-loop PID controller both in economic performance and in the size of its operating window. Furthermore, the QDMC controller achieved no less than 99.3% of the optimum total profit for both disturbance sets examined.