摘要
为了提高基于群体智能的粗糙集最小属性约简算法的求解质量和计算效率,提出一个结合长期记忆禁忌搜索方法的粒子群并行子群优化算法.并行的各子群不仅具有禁忌约束,而且包含多样性和增强性策略.由于并行的子群共同陷入局部最优的概率小于一个粒子群陷入局部最优的概率,该算法可提高获得全局最优的可能性,并减少受初始粒子群体的影响.多个UC I数据集的实验计算表明,提出的算法相对于其他的属性约简算法具有更高的概率搜索到最小粗糙集约简.因此所提出的算法用于求解最小属性约简问题是可行和较为有效的.
In order to improve the solution quality and computing efficiency of rough set minimum attribute reduction algorithms based on swarm intelligence,a parallel particle sub-swarm optimization algorithm with long-memory Tabu search capability was proposed.In addition to the taboo restriction,some diversification and intensification schemes were employed.Since parallel sub-swarms have a lower probability of simultaneously getting trapped in a local optimum than a single particle swarm,the proposed algorithm enhances the probability of finding a global optimum and decreases the influence of initial particles.Experimental results on a number of UCI datasets show that the proposed algorithm has a higher probability of finding a minimum attribute reduction in rough sets compared with some existing swarm intelligence based attribute reduction algorithms.Therefore,the proposed algorithm is feasible and relatively effective for the minimum attribute reduction problem.
出处
《智能系统学报》
2011年第2期132-140,共9页
CAAI Transactions on Intelligent Systems
基金
国家自然科学基金资助项目(60805042)
福建省自然科学基金资助项目(2010J01329)
关键词
属性约简
粗糙集
禁忌搜索
粒子群优化算法
并行子群
attribute reduction
rough set
tabu search
particle swarm optimization
parallel particle sub-swarm
