摘要
粒子群优化算法是一种解决非线性、不可微和多峰值复杂优化问题的优秀算法,但该算法在进化后期容易出现速度变慢以及早熟的现象;BP神经网络的学习算法是基于梯度下降这一本质的,因此存在着容易陷于局部极小值,收敛速度慢,训练时间长等问题。针对上述现象,对粒子群优化算法进行了增强粒子多样性和避免种群陷入早熟两个方面的改进,并提出了一种基于改进算法的粒子群神经网络算法,最后通过在IRIS数据集上进行的仿真实验验证了改进的有效性。
The particle swarm optimization arithmetic is an excellent optimization arithmetic that can solve the non-linear, un-fluxionary and multi-peak value optimizing problems. But in the process of looking for the excellent result, it is easily appear the phenomenon of speed becoming slow and precocious. The learning arithmetic of back propagation is base on the essence of grads descending, so there are inevitably problems of it is easy to get into partial least extremum, slowly constringency speed, long training time and so on. Improve the arithmetic at intensifying multiformity of particles and escaping the precocity of swarm, and put forward a particle swarm optimization neural network arithmetic based on the improved arithmetic. Prove the validity of the improving by the simulant experiments on the IRIS database.
出处
《计算机工程与设计》
CSCD
北大核心
2008年第11期2890-2892,2896,共4页
Computer Engineering and Design
关键词
粒子群优化
神经网络
群智能
BP算法
粒子多样性
particle swarm optimization
neural network
swarm intelligence
back propagation arithmetic
multiformity of particles
作者简介
何佳(1981--),女,河北唐山人,硕士,助教,研究方向为多媒体数据库技术;E—mail:hejiamail@yahoo.com.cn
陈智慧(1979-),女,河北唐山人,硕士,助教,研究方向为面向对象程序设计、数据安全;
杨迎新(1972~),女,河北唐山人,博士研究生,副教授,研究方向为智能数据库技术。