摘要
粒子群优化(PSO)的K-Medoids进化聚类算法中初始种群是随机产生的,导致选择的初始中心点有可能位于同一类簇中。为提高聚类准确性,提出一种采用递减概率化初始点选择的PSO与K-Medoids结合新算法。根据样本的分布密度设置对应的选择概率,并由轮盘赌策略依次选择中心点,使获得的中心点位于密度较高区域且在不同的簇中,同时又实现了初始种群的多样性。在人工和UCI真实数据集上的实验结果表明,改进后的算法有更快的收敛速度,提高了聚类准确率和稳定性。
The initial population in current K-Medoids evolutionary clustering algorithm based on PSO is generated randomly, which may lead to the same cluster of initial mediods. To solve this problem, a new algorithm combining PSO and K-Medoids based on decreasing-probability initial mediods selection is proposed. After setting the selected probability of each sample according to its density, they are selected in turn by roulette wheel selection in order to lo- cate them in higher density regions and different clusters, and the method can achieve the population diversity. The simulation results of both manual and UCI real datasets demonstrate that the proposed algorithm has a faster convergence speed and higher accuracy, and performs more stable.
出处
《计算机仿真》
CSCD
北大核心
2014年第9期314-318,共5页
Computer Simulation
关键词
递减概率化
粒子优化替换
粒子群优化
Decreasing-probability
Optimizing replace of particles
Particle swarm optimization (PSO)