Generalised Predictive Control

Generalised Predictive Control (GPC) is one of the control design methods that are able to cope with constraints on-line. It is based on ARIMAX model of the controlled process. This gives a possibility to estimate the model on-line using standard RLS methods.

Stable predictive control uses the main idea of GPC but requires that some internal signals are polynomials and not sequences. Computationaly more intensive than GPC but provides beyond stability some more interesting features.

Two directions in stable predictive control (SPC) are investigated: Unification of many known SPC methods is given in (TR MF9902,CSD'00,ROCOND'00) is given based on the location of the closed-loop poles. The unified predictive controller is able to minimise several cost functions: finite and infinite horizon settings, generating the same closed-loop actions as given nominal linear controller.

The second direction tries to minimise the extent of performance degradation in the case of constrained control (TR MF0002, CSD 2003). This is obtained by generating the desired output trajectory as the output of the nominal unconstrained closed-loop system and minimising the surface between the constrained and unconstrained trajectories. Both simple and stabilising methods are proposed, the latter consists in serial use of GPC and linear controller.



Technical Reports

Last Modification: