Genetic Algorithm Based Controls
Dissertations and Theses
Control theory has experienced a great deal of advancement over the past several decades. The control of linear time-invariant systems with known dynamics has been thoroughly explored. When faced with uncertainties in the system's dynamics, techniques such as H-infinity optimal control have proven effective and robust . A variety of methods have been developed to control systems with nonlinear dynamics. These nonlinear control methods work well if the system is linearizeable; however, many linear approximations to nonlinear systems yield inconclusive models which do not provide a means to control the system . The control of unknown systems is currently a popular research topic, employing techniques such as system identification, neural networks, and fuzzy logic. These methods have been extended to adaptive techniques for the control of time-varying systems. Still, a single method sufficiently robust to apply to such a broad class of systems has yet to be developed.
In recent years, researchers have become increasingly interested in the use of Genetic Algorithms as a means to control various classes of systems. Genetic Algorithms are robust search techniques based on the principles of evolution. Extensive research has been performed exploiting the robust properties of Genetic Algorithms and demonstrating their capabilities across a broad range of problems. These evolutionary methods have gained recognition as general problem solving techniques in many application, including function optimization, image processing, classification and machine learning, training of neural networks, and system control. This study focuses on the use of Genetic Algorithms in the control of unknown systems, and in particular, the stability of those systems.
This study presents a method of adaptive system control based on genetic algorithms. The method consists of a population of controllers evolving towards an optimum controller through the use of probabilistic genetic operators. A brief overview of genetic algorithms is first given. The remainder of the paper identifies the problems associated with genetic algorithms controllers, and addresses the key issue of stability. A theoretical analysis of the proposed genetic algorithm controller shows the population converges to stable controllers under fitness-proportionate selection pressure. The minimization of the effects of instability is also discussed.
Genetic algorithms are stochastic search techniques that guide a population of solutions using the principles of evolution and natural genetics. In recent years, genetic algorithms have become a popular optimization tool for many areas of research, including the field of system control and control design. Significant research exists concerning genetic algorithms for control design and off-line controller analyses. However, little work has been done with on-line genetic algorithm controls primarily because of the problems associated with instability in early stages of the controllers evolution. Also, until now the stability of controllers based on genetic algorithms has not been researched in detail.
This study presents a genetic algorithm controller that consists of a population of controllers, each of which control the system for a specified time period. A brief overview of genetic algorithms and a history of genetic algorithms in system controls is provided, followed by a detailed discussion of the developed controller. The scope of the research encompasses an analysis of the stability of the resulting control system with respect to the convergence of the genetic algorithm. The results of several simulations are given in support of the developed theory. Also, issues related to practical considerations of the genetic algorithm controller are discussed for future development.
Genetic algorithms are stochastic search techniques based on the principles of evolution. Extensive research has been performed exploiting the robust properties of genetic algorithms and demonstrating their capabilities across a broad range of problems. These evolutionary methods have gained recognition as general problem solving techniques in many applications, including function optimization, image processing, classification and machine learning, training of neural networks, and system control.