Economic Model Predictive Control for uncertain systems

Abstract

Usually, in the chemical process operation, economic and control problems are solved separately, in two layers. The economic optimization layer is known as Real Time Optimization, RTO. Based on the rigorous process model, the RTO routine calculates the steady-state values that maximize the profit. The solution of the RTO produces the targets that are sent to the Model Predictive Controller, MPC. The MPC layer takes this information and solves a second optimization problem. It has to put the process outputs as close as possible to the targets, considering the control constraints and a linear dynamic process model. The presence of the rigorous model constraint introduces nonlinearities in the optimization problem and increases the computational load. Then, the routines must be solved at different frequencies. The MPC optimization routine is typically a QP problem, it is solved in a minute or second scale, while the RTO, that solves a nonlinear problem, requires more time. In a frequent disturbance scenario, this operational structure produces a lag between the real economic optimum and the target value that the controller is implementing. Then, the purpose of this research project is the integration of the economic and the control problems into a single optimization problem, in order to obtain an economic MPC with guarantee of stability. This new controller must be able to deal with the main drawbacks of the conventional strategy. In order to preserve the QP structure of the MPC, it is intended to study the insertion of a linear economic term in the objective function of a stable MPC. This term should not disrupt the closed-loop stability. Furthermore, considering that in real systems the linear prediction model is not accurate, the proposed approach will be extended to uncertain systems. (AU)