Research On Genetic Algorithm And Its Application In Thermal Process Modeling And Optimal Control

With the development of power industries, the scale and parameters of the generator units have been increasing with full automation. They bring forward higher demand on control quality of control system. The math model and control system design after modeling are the base and key point to process control system analyses, design, debug and obtain the higher control quality. Therefore, research on the method of system modeling and controller parameters tuning is very important. This thesis puts up the research aims at genetic algorithm and its application in modeling and optimal control.The main contents are made up of three parts. The first part researches the thermal process model identification based on genetic algorithm. The algorithm adopts adaptive crossover and mutation strategy with multi-operators, and improves the precision, search rate and ability of constringency. The algorithm adopts adaptive crossover and mutation strategy with multi-operators, and improves the precision, search rate and ability of constringency. The zero-pole transfer function is used to describe the process objective and the familiar thermal process models are classified. Thus the bugs can be avoided that are met while adopting general transfer functions, such as parameters'difference, cooperation and searching, and the adaptive genetic algorithm is used to thermal process identification. Through the simulation research, the model identification method based on genetic algorithm has ability to identify model, and it can obtain result with high precision. The second part researches on multi-objective optimization with genetic algorithm. It analyses and compares advantages and disadvantages of many typical algorithms. Based on these, two improved multi-objective optimization with genetic algorithm are proposed. Uniformity weight-sum based on non-dominated ranking method improves the idea of traditional weight-sum in two objectives optimization. The objective weight coefficients change symmetrically in a certain space and solve the problems of coefficients distribution. Meanwhile, the excellent results which are obtained in every weight coefficients are ranked based on non-dominated ranking method and they can distributing more uniformity on Pareto front. The improved Pareto multi-objective optimization with GA adopts non-dominated ranking, elitist preserve and niche approaches, and proposes individual vector module adaptive function as wash out rule. As an example, two duality functions are used to validate the two algorithms. The results indicate that both of these two algorithms obtain the Pareto front with distributing uniformity. Especially, as used niche technology, the improved Pareto multi-objective optimization with GA avoids the problem of local convergent and optimal solutions distribute more uniformity. The third part based on introducing PID control principle, the rules of controller parameters tuning and familiar tuning methods, PID controller parameters tuning based on...