whole_DiritaVito2002_thesis.pdf (42.68 MB)
Control system design applications with hybrid genetic algorithms
thesisposted on 2023-05-26, 22:46 authored by Dirita, Vito
This thesis investigates the hybrid application of stochastic and heuristic algorithms, in particular genetic algorithms (GA), simulated annealing (SA) and Greedy search algorithms for the design of linear and nonlinear control systems. We compare the rate of convergence, computational effort required (FLOPS) and ease of implementation. Where possible, results are compared with the more traditional control system design methodologies. Two specific practical applications include aircraft flight control systems, and a nonlinear example of an industrial bioreactor fermentation process. Stochastic algorithms (GA) and heuristic algorithms (SA, Greedy, Tabu search) are powerful search methods, capable of locating the global minimum or maximum (extremum) of multimodal functions. They operate without the need for function gradients and are robust to noisy data. The current research trend is directed towards the solution to constrained multiobjective optimization problems of multimodal functions which may result in a family of optimal solutions (i.e Pareto optimal set) and game theoretic approaches such as Nash and Stackelberg Equilibria. Genetic algorithms suffer from one particular drawback, the rate of convergence can be unacceptably slow if accurate solutions are sought. To overcome this deficiency, hybridization of genetic algorithms with fast local search procedures are often used. Two heuristic based search procedures are: greedy search and fast simulated annealing. We investigate three types of Hybrid algorithms: (i) genetic algorithms (GA), (ii) hybrid GA + simulated annealing (SA), and (iii) hybrid GA + greedy search. These methods are applied to solving off-line linear and nonlinear control problems which may otherwise have no direct analytical solution. In cases where solutions are obtainable using conventional methods, results are compared with hybrid algorithms. Robustness against modeling errors, nonlinearities, disturbances and parametric uncertainty will also be discussed. We investigate five specific design applications, these include: training radial basis function (RBF) neural networks, robust eigenstructure assignment (ESA), model reference adaptive control (MRAC), robust mixed H2/H00 design, and lastly fault detection and isolation (FDI). We show that hybrid algorithms can perform better, can handle a broader class of problems, and have fewer restrictions than conventional methods. Furthermore, stochastic and heuristic methods can directly deal with constraints.
Rights statementCopyright 2002 the Author - The University is continuing to endeavour to trace the copyright owner(s) and in the meantime this item has been reproduced here in good faith. We would be pleased to hear from the copyright owner(s). Thesis (Ph.D.)--University of Tasmania, 2002. Includes bibliographical references