theses_web.jpg

Thesis for the Degree of Master of Science

 

Year 1998
Jarkko Kangas

Optimal constraint handling of multivariable model predictive controller


Methods of constraint handling in model predictive multivariable controllers were examined in this thesis. The study focuses particularly on state variable constraints.

An essential characteristic of model predictive control methods is their ability to take into account process constraints. The formulation of control objectives and constraints in discrete time domain results in a constrained optimization problem from which the exact optimal control sequence can be solved with methods of mathematical programming. However, in large multivariable systems the complexity of the optimization problem is high and it is not always possible to execute a rigorous mathematical programming algorithm within one control cycle. Process disturbances and model mismatch can also give rise to situations when a feasible solution, i.e., a solution satisfying all the constraints does not exist.

In the literature part, the formulation of constraints and methods of handling infeasibilities in model predictive controllers were considered. Mathematical programming algorithms were also examined in the context of predictive control.

In model predictive control applications the control calculation can be speeded up and the problems caused by infeasibilities can be avoided by simplifying the constrained optimization calculation. In the experimental part of the thesis, a simplified state constraint handling method was developed based on an industrial multivariable predictive controller. The objective was to examine possibilities to improve the constraint handling capabilities of the controller in order to decrease the effort needed in the tuning phase.

The controller developed was simulated by using a model of a binary distillation column and the results were compared with those of the original controller. The constraint handling of the new controller was found to perform remarkably better after set point changes even though the same tuning parameters were used in both controllers. The possibilities of applying the method to the industrial controller look good.


This info last modified 22 Nov 2017 - contact webmaster