Last challenges in Model Predictive Control applied in industry demand to focus in optimization process. New methods for substitute the classical ‘linear programming’ approximation or trial and error procedures in optimization process and predictive controller tuning are needed.
Multiobjective optimization strategy so-called Physical Programming allows controller designers a flexible way to express design preferences with a ‘physical’ sense. For each objective (settling time, overshoot, disturbance rejection, etc.) preferences are established through categories as desirable, tolerable, unacceptable, etc. assigned to numerical ranges (the same for plant optimization (productivity, energy costs, residuals, etc.). The problem is translated to a unique objective optimization but normally as a multimodal problem. This project will show how to convert to multiobjective optimization the problems of parameter tuning in multivariable predictive controller and controller set point selection for process optimization. The proposal will try to solve them with Physical Programming and Evolutionary Algorithms due to the multimodal characterization of the cost functions obtained. Finally, the implementation of a multiobjective optimization software able to communicate with commercial software packages of predictive control is in mind.
KEYWORDS: Multi-objective optimization, evolutionary algorithms, plant economics, decision making, multivariable control, model predictive control