On Problems Of Robust Control Based On Genetic Algorithm

Development for feedback control theory is divided into three phases, i.e. classical control theory, modern control theory and robust control theory. Classical control theory is applied to solving of complex control problems. The cut and try method is, however, the main designing way of control systems, which can not provide an optimization method. Modern control theory mainly applies the method of state space to study the state movement rule of control systems, and obtain the optimization design. At the same time, there is inevitable uncertainty in the practical plant; therefore it is difficult to entirely apply modern control theory and method to practical process. In order to remedy the shortage of modern control theory, robust control theory sufficiently considers all kinds of uncertainty of controlled object in the analyses and design phase. And analyses and designs o f controllers are obtained based on uncertainty and non-precise model.The latest results of the robust control, especially in the aspects of the robust stability theory, the problems of H_∞, control, μ analysis and μ synthesis etc. of linear systems possiss important significance in the last ten years. However, when the results are applied to practice, there are some problems such as the order of the controller is high, the structure is complex, the process of the resolving is fussy, the conservatism of the system is not strong and etc. Therefore, the genetic algorithm (GA) introduced to robust control problems in the dissertation. The design and synthsis problems of control systems are discussed. The robust control problem is transformed into an optimization problem with restrictions. A method is presented by using genetic algorithm, which has comprehensive optimization searching ability to resolve these problems. The main work of the dissertation is generalized as follows.An improved real adaptive genetic algorithm (RAGA) is presented. In practical applications, compared with binary valued coding, the nature and intuitionistic method of real coding has the merit of high precision and good stability. However, it also has the shortcomings: the search is inefficient and it is easy to premature convergence. The adaptive crossover probability and adaptive mutation probability are proposed, which consider the influence of every generation to algorithm and the effect of different individual fitness in every generation.