本文提出一种基于统计学习理论优化感知器的遗传方法。该方法将遗传算法和神经网络相结合,通过统计学习理论指导遗传算法优化分类器的过程,避免了传统的感知器分类的偏向性、连接权的局部收敛性、误识率高等弱点;借助于遗传算法全局寻优的特点,使改进后的算法,具有自进化、自适应能力,以及很好的数据推广性能和抗干扰性,提高了神经网络的整体性能,与标准的SVM算法相比,具有更广阔的应用范围。 This paper presents a genetic algorithm based on statistical learning theory to optimize perceptrons. The method combines genetic algorithm and neural network, and guides the genetic algorithm to optimize the classifier through statistical learning theory. It avoids the bias of the traditional perceptron classification, the local convergence of connection rights and the high false positive rate. With the aid of The global optimization of genetic algorithm makes the improved algorithm have the ability of self-evolution, self-adaptability, good data promotion and anti-interference, and improve the overall performance of neural network. Compared with the standard SVM algorithm, Has a broader range of applications.