没有合适的资源?快使用搜索试试~ 我知道了~
首页提升全局最优搜索能力的粒子群优化算法研究
提升全局最优搜索能力的粒子群优化算法研究
3 下载量 166 浏览量
更新于2024-08-26
收藏 329KB PDF 举报
"增强全局最优粒子搜索能力的改进粒子群算法" 本文是关于“增强全局最优粒子搜索能力的改进粒子群算法”的一篇研究论文,发表于2015年的IEEE第12届国际电子测量与仪器会议(ICEMI'2015)。作者团队来自中国电子科技大学自动化工程学院,包括张伟、石义兵、马东和刘国珍。 粒子群优化(PSO)是一种基于群体智能的优化算法,它模拟了鸟群寻找食物的过程来解决复杂的优化问题。然而,PSO在迭代过程中存在一个问题:一旦某个粒子成为全局最优粒子,它的运动多样性会显著减弱,搜索能力也会大幅下降。这一现象限制了PSO的收敛速度和全局搜索性能。 为了解决这个问题,论文提出了一种新颖的改进策略,即通过增强全局最优粒子的搜索能力来提高PSO的收敛速度。具体做法是,增大全局最优粒子的惯性系数,使其比其他粒子具有更大的探索空间,同时调整全局最优粒子的搜索方向,以避免陷入局部最优。这种策略旨在保持全局最优粒子的活力,防止其过早地被锁定在一个可能的解上。 为了验证所提出的改进算法的有效性,研究者使用了四个基准函数进行测试。这些基准函数通常用来评估优化算法的性能,包括多元函数的全局最小值搜索。通过对比分析,结果表明,改进后的粒子群算法在保持搜索效率的同时,能更好地避开局部极小值,提高了全局寻优能力。 这篇论文为粒子群优化算法提供了一个创新的解决方案,增强了全局最优粒子的搜索能力,有望应用于各种需要全局优化的复杂问题,如工程设计、机器学习模型的参数优化等场景。这项工作对于理解和改进群体智能算法,以及提升优化问题的求解效率具有重要意义。
资源详情
资源推荐
2015 IEEE 12th International Conference on Electronic Measurement & Instruments ICEMI’2015
Modified particle swarm optimization algorithm by enhancing search ability of
global optimal particle
Zhang Wei, Shi Yibing, Ma Dong, Liu Guozhen
School of Automation Engineering, UESTC
No.2006, Xiyuan Ave., West Hi-Tech Zone, 611731 Chengdu, China
Email: weizhang@uestc.edu.cn
Abstract –Firstly, the laws of the Instantaneous movement of
different particles were analyzed in geometric view in this paper.
The important conclusion “In the process of Particle Swarm
Optimization (PSO) iteration, the movement diversity will be
seriously weaken and the search ability will be greatly reduced
as soon as a particle becomes the global optimal particle” was
drawn. Then, for the purpose of improving the convergence
speed of PSO, a novel improved strategy based on the search
ability enhancement of the global optimal particle which is by
means of making the inertia coefficient of the global optimal
particle bigger than other particles and adjusting the search
direction of the global optimal particle. Four benchmark
functions were used to test the proposed improved PSO
algorithm, the standard PSO algorithm and the PSO algorithm
with the leader. The variance analysis of statistic theory is used
to compare the performance of the three algorithms.
Experiments show that the proposed algorithm converges faster
in the optimization of single extreme value and multiple extreme
values without severe oscillation.
Keywords –Geometric analysis, PSO, inertial coefficient, search
ability enhancements, least significant difference multiple
comparisons
I. INTRODUCTION
J.Kennedy and R.Eberhat carried out simulation
research on birds’ predatory behavior and proposed the
particle swarm optimization for the first time in their
famous paper in 1995
[1]
. In order to bet better control of
particle swarm’s search behavior, Y.Shi and
R.Eberhart made a significant improvement on the PSO
in reference [2]: inertia weight coefficients ω, of which
the choosing method was introduced in reference [3],
have been added to the velocity iteration formula, this
improvement lead to the birth of the standard PSO
which is widely used at present. In order to enhance
global search capability, the mutation operation of
genetic algorithm is introduced into standard PSO
[4-6]
.
Meanwhile, the global optimal
particle has also been studied further by some scholars.
Standard PSO takes particle which has the best fitness
in particle swarm as the global optimal particle. In
reference [7], an additional particle which is the geometric
center of particle swarm has been brought into the
calculation process of global optimal particle. PSO search
process is divided into exploration search and development
search in reference [6], and it has been pointed out that
global convergence capability of PSO can be improved by
giving the global optimal particle an inertia coefficient far
more lager than which is given to any other particles during
search processes. Reference [8] brings gravity search
algorithm into standard PSO and fitness value of a particle
is added its position as a weight value, global optimal
particle is regarded as gravity center and global
convergence capability of PSO can be improved. In
reference [9], weighted random values are added to
equilibrium particle’s original velocity vector, self-learning
vector and social learning vector in order to enhance the
global search capability of PSO. In reference [10], the
learning process of particle swarm is considered as a chaos
system to prevent algorithm falling into local optimum.
This paper discusses instantaneous motion rule of
particles in PSO from the perspective of geometry and
obtains important factors that affect convergence
velocity of PSO. Based on an important conclusion that
a particle’s motion diversity will be severely weakened
and its search capability will be greatly reduced as soon
as it is turned to a global optimal particle during the
iteration process of standard PSO, a kind of PSO whose
global optimal particle’s search capability is enhanced
is proposed.
II. GEOMETRIC ANALYSIS OF
INSTANTANEOUS MOTION RULE
OF PARTICLES
Every particle in standard PSO iterates according
to formula (1) and formula (2). Formula (1) is the core
formula of iteration of PSO, and it decides the motion
velocity and direction of particle iteration.
1
11 2 2
()()
tt t t
ii ii i
v v c r pbest x c r gbest x
(1)
11ttt
iii
x
xv
(2)
The position and velocity in formula (1) and
formula (2) are both vectors,
1, 2, ,in
, ω is inertia
coefficient, c
1
and c
2
are accelerated factors, usually
12
2cc
. r1 and r2 are random values that are
evenly distributed in section from 0 to 1.
Assuming that the solution-search space Ω is a
Hilbert space, then formula (1) can be viewed as a
linear combination of 3 vectors of Ωspace. The update
velocity vector
1t
i
v
of any particle i is the linear
combination of the particle’s current velocity vector,
distance vector
()
k
ii
p
best x
that is between current
position and historical optimal position and distance
____________________________________
978-1-4799-7071-1 /15/$31.00 ©2015 IEEE
451
下载后可阅读完整内容,剩余6页未读,立即下载
weixin_38680393
- 粉丝: 6
- 资源: 912
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
最新资源
- WebLogic集群配置与管理实战指南
- AIX5.3上安装Weblogic 9.2详细步骤
- 面向对象编程模拟试题详解与解析
- Flex+FMS2.0中文教程:开发流媒体应用的实践指南
- PID调节深入解析:从入门到精通
- 数字水印技术:保护版权的新防线
- 8位数码管显示24小时制数字电子钟程序设计
- Mhdd免费版详细使用教程:硬盘检测与坏道屏蔽
- 操作系统期末复习指南:进程、线程与系统调用详解
- Cognos8性能优化指南:软件参数与报表设计调优
- Cognos8开发入门:从Transformer到ReportStudio
- Cisco 6509交换机配置全面指南
- C#入门:XML基础教程与实例解析
- Matlab振动分析详解:从单自由度到6自由度模型
- Eclipse JDT中的ASTParser详解与核心类介绍
- Java程序员必备资源网站大全
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功