用改进的遗传算法训练神经网络构造分类器

针对基本遗传算法存在容易早熟和局部搜索能力弱等缺陷，提出了改进的遗传算法，引入交叉概率和变异概率与个体的适度值相联系，改进了操作算子，而且在交叉操作后又引入模拟退火机制，提高遗传算法的局部搜索能力。同时，用改进的遗传算法和基本的遗传算法训练神经网络构造分类器，实验结果表明，改进的遗传算法在最好个体适度值和最好分类准确性等方面性能更好。

Constructing classifier of neural networks using improved genetic algorithms

An Improved Genetic Algorithms(IGA) was presented. IGA adopted crossover probability and mutation probability decided by individual's fitness, introduced simulated annealing methods after crossover, and improved operators of Simple Genetic Algorithms(SGA), in order to avoid drawbacks such as prematurity and bad local search ability etc of SGA. In this paper, classifiers of neural networks were constructed using IGA and SGA. Experiment results show that IGA performs better than SGA on the best fitness and the best classifying veracity.