摘要
提出了一种基于免疫遗传算法(IGA)的BP神经网络设计方法.该算法在遗传算法(GA)的基础上引入生物免疫系统中的多样性保持机制和抗体浓度调节机制,有效地克服了GA算法的搜索效率低、个体多样性差及早熟现象,提高了算法的收敛性能.为了解决BP神经网络权值随机初始化带来的问题,用多样性模拟退火算法(SAND)进行神经网络权值初始化,并给出了算法详细的设计步骤.仿真结果表明,同混合遗传算法相比,该算法设计的BP神经网络具有较快的收敛速度和较强的全局收敛性能.
A new method of designing BP neural networks based on immune genetic algorithm （IGA） was proposed. The mechanisms of diversity maintaining and antibody density regulation exhibited in a biological immune system were introduced into IGA based on genetic algorithm （GA）. The proposed algorithm overcame the problems of GA on search efficiency, individual diversity and premature, and enhanced the convergent performance effectively. In order to solve the problem of random initial weights, simulated annealing algorithm for diversity was used to initialize weight vectors, and the detailed design steps of the algorithm were given. Simulated results show that the BP neural networks designed by IGA have better performance in Convergent speed and global convergence compared with hybrid genetic algorithm.
出处
《北京科技大学学报》
EI
CAS
CSCD
北大核心
2006年第10期997-1000,共4页
Journal of University of Science and Technology Beijing
关键词
BP神经网络
免疫遗传算法
模拟退火算法
全局收敛性
BP neural network
immune genetic algorithm
simulated annealing algorithm
global convergence
