含规则数优化的多目标属性约简进化算法.pdfVIP

小型微型计算机系统 2016 年8 月第8 期 JournaI of αúnese Computer Systems Vol. 37 No.8 2016 含规则数优化的多目标属性约简进化算法 洪华剑，叶东毅 {福州大学数学与计算机科学学院，福州 350108) E-mail: 蚓hongl蜘@qq. ∞m 摘 要:在粗糙集理论中，最小属性约简未必是对应决策规则数量少的约简.为了从属性数和规则数两个维度消除数据表的冗 余信息，提出一种以最少提取规则数和最少属性个数同时为优化目标的多目标属性约简问题及其相应的多目标遗传进化求解 算法.该算法在 NSGA2(Non-dominated S创世ng Genetic Algorithm 2) 算法的基础上，以多目标之间的支配关系确定种群个体优 先级，并有针对性地引入了精英保留策略、分散进化策略和去重策略.实验结果表明，该算法能够有效地求解本文提出的多目标 属性约简问题，其中的分散进化策略、去重策略和精英保留策略增强了进化种郡的多样性和收敛性.与 NSGA2算法比较，本文 算法能获得更多的非支配多目标约简，具有更优的求解能力. 关键词:属性约简;粗糙集;规则提取;多目标优化;遗传算法 中固分类号: TP18 文献标识码:A 文章编号:1创lO-1220( 2016 )08-1707 -05 An Evolutionary Algorithm for Multi-objective AUribute Reduction Involving Optimization of the Number of 胁cision Rulω HONG Hua-jian ,YE Dong-yi ( College of M础ematics and 臼呻阳Science ,FuzI刷刷versity ， Fl础。u 350108 ， αùna) Abstract: In the rough set theory ,the minimum attribute reduction does not necessarily correspond to 由e reduction of minimum deci- sion rules. In order ωeliminate the redun也ot information in da能set from the number of attributes and decision rules pers户口ive ， we propose a mwti-objective a伽ibute reduction 萨。blem as well as its corresponding multi-o时ective genetic evolutionary aIgorithm ,which aims ωget the le副 number of decision rules and a时butes simul能neously. Ba能d on NSGA2 (Non-dominated Sorting Genetic Algo- ri由m2) ，也.e aIgorithm determines individuaI priority in a population by dominance relationship between multiple objective ,and con- 回puntaIly in缸。duces elitism strate町， decentralized evolt皿on strategy and removing duplication strategy. 币le experimentaI results show 阳t 也e 严oposed aIgorithm can effectively solve the proposed problem of multi-o均ective attribute reduc往on. Moreover ,elitism strategy ， decentr对ízed evolution stra能gy and removing duplication s田.te盯归由e aIgor