Thesis supervisor: József Tar
Location of studies (in Hungarian): Óbuda University Abbreviation of location of studies: ÓE
Description of the research topic:
The mathematical structure of the "Receding Horizon Controller" corresponds to the realization of the "Model Predictive Controller" that makes design for a finite horizon approximated as a discrete time-grid by the use of Lagrange's Reduced Gradient Method often referred to as Nonlinear Programing. In this approach a compromise can be realized between contradictory requirements formulated in cost function contributions, and the "strict conditions", i.e. the abilities or the "model" of the dynamic system under control are taken into consideration in the form of constraint equations. The effects of modeling imprecisions and the unknown external disturbances are taken into consideration by the application of frequent redesign of the horizion on the basis of actually measured system states.
In the practice neither precise system models are available, nor realistic possibility exists for the estimation of the full state variable. For tackling such problems the prevailing solutions use Lyapunov's 2nd method to design adaptive controllers. In general it can be stated that the integration of the mathematical structures of the optimal and the adaptive controllers is not an easy task.
At the Óbuda University recent attempts were done to replace Lyapunov's 2nd method in the adaptive control by a simpler iterative approach based on Banach's Fixed Point Theorem proved in 1922. It was found that the mathematical structure of this approach easily can be combined with that of the optimal controllers. Furtehrmore, attempts were done to replace Lagrange's method with fixed point iteration in the optimization, too.
The aim of the here outlined research is the continuation of these preliminary steps by systematically investigating the effects of horizon length, grid resolution, the realtive order of the control, the effects of nonsingular and singular dynamic system models, and that of the state estimation based on approximate dynamic models due to being in the lack of satisfactory number of sensors.