Hybrid Evolutionary-Algebraic Approaches for Controller Design
The aim of this project is to research the possibilities for applying modern heuristic methods in the problems of low order and fixed structure H-infinity controller design. Many advanced controller design techniques, like H-inf, H2 and mu synthesis, have limited application in the industry, one of the reasons for which is the often high order of the designed controller (plant + shaping filters), which limits its practical applicability. The widely used order reduction (either of the plant or of the controller) approach does not guarantee good performance of the low-order controller on the high order plant. On the other hand the direct design of low order controllers is a non-convex problem and thus difficult to solve. In this project the problems of fixed order and fixed structure H-infinity controller design are considered. The tasks of this project are: Explore the possibilities of combining evolutionary algorithms approach with the standard controller design techniques in a way that exploits the good characteristics of both of them, thus allowing computation time reduction and improved reproducibility of the results. Research of the application possibilities for cooperative coevolutionary methods for simultaneous search for controller (K) and scaling (D) during mu synthesis. Compare the evolutionary based methods with recently developed local optimization methods for non-smooth and non-convex optimization. Determine the advantages and areas of applicability of each of the approaches. PoWe06, Po06, PoFaWe05, Po05
|