一种利用Gauss变异优化BP神经网络的方法

摘要

In order to improve the traditional BP algorithm,whose convergence speed is low and which is easy to fall into local minimum,an improved training method of BP neural network is proposed,in which the genetic algorithm is combined with the improved BP algorithm. The concept of population in genetic algorithm was introduced. The error of the best individual was selected as the minimum error. Two small random numbers generated by Gauss mutation was taken as the learning rate and the impulse coefficient in BP algorithn to realize the dynamic adjustment of the two parameters and weight optimization of BP net-work. Experimental results show that the method improves the convergence speed of the neural network effectively,and has an obvious superiority in training time and a good practicability.%为了克服传统BP算法收敛速度慢和局部极小点等问题，提出了一种改进的BP网络训练方法，将改进的BP算法和遗传算法相结合。首先引入遗传算法中群体的概念，选取最好个体中的误差作为最小误差，其次利用Gauss变异生成的两个小随机数作为BP算法中的学习率和冲量系数，实现对两个参数的动态调整，以达到对BP网络的权值优化的目的。实验结果表明，该方法有效提高了BP网络的收敛速度，在训练时间方面具有明显的优越性，具有较好的实用性。