一种改进的基于粒子群的聚类算法

Improved clustering algorithm based on particle swarm optimization

针对K-means对初始聚类中心敏感和易陷入局部最优的缺点,提出了一种改进的基于粒子群的聚类算法。该算法结合基于密度和最大最小距离法来确定初始聚类中心,解决K-means对初始值敏感的问题;利用粒子群算法全局寻优能力强的优点,避免K-means陷入局部最优。通过对样本集各维属性的规范化处理,惯性权值采用凹函数递减,计算相异度矩阵,引入用群体适应度方差,进一步优化混合算法。实验结果表明,该算法具有更高的准确率和更强的收敛能力。

This paper proposed an improved clustering algorithm based on particle swarm optimization that was aimed to re- solve the K-means algorithm shortcoming of sensitiving to the initial clustering center and easiness to fall into local optimum. The improved algorithm combined density-based and maximum minimum distance method to determine initial clustering center, solved problem of the K-means was sensitive to the initial clustering center. The advantages of PSO＇ s strong global optimization ability were used to avoid K-means falling into local optimum. By normalizing each attribute of the sample set, inertia weight was decreased by concave function, calculated the dissimilarity matrix and introduced particle swarm＇ s fitness variance to opti- mize the hybrid algorithm further. The experimental results show that this algorithm has higher accuracy and stronger conver- gence capability.