文献类型：期刊文献

中文题名：基于粒子群算法的函数复杂度分类法

英文题名：A Classification Model to Classify a Given Function Based on Its Complexities Using Particle Swarm Optimization

作者：刘静秋[1];杜文莉[1];张飞[1]

机构：[1]华东理工大学化工过程先进控制和优化技术教育部重点实验室,上海200237

年份：2020

卷号：27

期号：8

起止页码：1337

中文期刊名：控制工程

外文期刊名：Control Engineering of China

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD_E2019_2020】；

基金：国家重点研发计划项目(2016YFB0303401);国家杰出青年科学基金(61725301);国家自然科学基金青年项目(61703163,21506050)。

语种：中文

中文关键词：粒子群优化算法;函数复杂度;复杂度分类

外文关键词：Particle swarm optimization;function complexity;classification model

摘要：不同优化问题在搜索空间内具有不同的空间形态,呈现出不同的函数复杂度特性。针对各种不同复杂度的函数,也各自存在其适宜的求解方法。但评价函数的复杂程度存在一定难度,观察到粒子群算法在优化给定函数时,其种群的动态规律性与函数的复杂程度相关,由此提出了基于粒子群算法的函数复杂度分类法。基于各个粒子的速度矢量定义了“局部收敛趋势度”指标,以实时量化种群的进化特征;观察种群进化特征数据,构建了经验式函数复杂度分类模型。应用大量测试函数对所提函数复杂度(单峰/多峰特性)分类模型的准确性进行了验证,错误率为2.516%。
Every optimization function has different landscapes within its search space.The differences can be seen as different function complexities.And functions with various complexities have their best suitable methods to enhance the optimization effects.However,it is difficult to evaluate how complex a given function is.It is found that during the optimization process of particle swarm optimization(PSO)algorithm,particles behave regularly in some way,and this regular pattern is closely related to the complexity of the optimized function.Based on that,a classification model is proposed to classify the given objective functions.Firstly,in order to quantify the evolutionary characteristics of the population in real time,the"local convergence trend degree"index based on the velocity vector of each particle is defined.Then,observing the data of population evolution characteristics,an empirical function complexity classification model is constructed.With lots of benchmarks,we have empirically analyzed the proposed classification model is.And the test results confirm the accuracy of the proposed classification model(error rate:2.516%).