The Studies On Some Improvements Of The GA And Their Applications In SVM

Genetic Algorithm is a kind of evolution algorithm. In recent years, not only a system of algorithm is formed relatively perfect, but the scope of application is extended greatly. At the some time, the genetic algorithm is improved constantly and some improved method is used to overcome the premature convergence phenomenon. The study of this paper is based on the improved method,some innovations are proposed.we do the following work:1. An iterative self-adaptation formula for the crossing rate and variance rate is proposed. To our knowledge, it is a new one. By substituting the crossing rate of the current generation for that of the first generation, which is given initially, in the iterative self-adaptation formula, a new iterative self-adaptation formula for crossing rate is given. Using the same technique to modify the iterative self-adaptation formula for variance rate,yeilds a new iterative formula for variance rate. This modification to the iterative self-adaptation formulas improves the searching efficiency of the genetic algorithm.2. In the previous non-linear fitness function that varies in the evolutionary process, there exist a estimated maximum generation. By replacing it with its counterparts in the current generation, a new non-linear fitness function that varies in the evolutionary process is derived. This modification improves the ruuing efficiency of the genetic algorithm.3. The modified genetic algorithm proposed in this thesis is employed to optimize the parameters of hybrid kernel function of support vector machine. And the support vector machine based on the modified genetic algorithm is used to estimate the software cost. Compared to the support vector machine based on the other genetic algorithms, the one in this thesis is better in terms of the accuracy of estimation.