| Summary: | 碩士 === 國立屏東科技大學 === 工業管理系所 === 101 === Genetic algorithms are processes following the biological evolutionary rules to search optimal solutions. Because genetic algorithms are based on simple principles and easily written into computer programs, the kind of algorithms becomes a popular tool to solve integer programming problems. The parameter setting is the most important issue for the performance of searching process of a genetic algorithm. Since some parameter values of genetic algorithms range in continuous intervals, the response surface method seems to be the best choice to design high performance parameters for a genetic algorithm. However, the surface response method is not popular when one designs the parameters for genetic algorithms. The reason is that parameters, the number of population and the number of iterations, are unbounded and the resulted value of objective function and the values of these two parameters are in the direct proportion. If one considers the parameters of genetic algorithms as the control factors and the resulted value of objective function as the response value, the response surface method will not stop iterating to increase the values of these two parameters. In addition, designing large numbers for these two parameters will result long computation time and that will make an algorithm inefficient. Therefore, this paper introduces a method to define the control factors instead of considers each of parameters of genetic algorithms as a control factor of the response surface method. Then, the response surface method can be used to design the parameters of genetic algorithms. The performance of the parameter setting from the response surface method will be proved by using a large scale integer programming, the Taiwan High Speed Rail Scheduling problem. The result shows that the parameter setting from the response surface method is better than the one from the popular Taguchi method.
|
|---|
