【摘要】 为了解决传统的K-means聚类算法全局优化性差,容易陷入局部最优的问题,用具有全局自适应优化特点的遗传算法与K-means算法结合来改善聚类效果.在此基础上提出了基于余弦因子改进的混合聚类算法(SGKM),在交叉和变异操作时用基因余弦因子(GCOS)进行个体控制,确保差的个体不会被引入下一代,并采用交叉和变异概率的自适应控制,结合了K-means算法的高效局部搜索和遗传算法的全局优化能力.实验结果表明,与其他基于K-means算法改进的聚类算法相比,SGKM算法能获得更小的簇内距和更大的簇间距,且数据对象的分类准确率有一定的提高.应用SGKM算法进行聚类不易受到不良个体的干扰,可以有效地改善聚类效果.更多还原

【Abstract】 To solve the problem of the traditional K-means clustering algorithm’s weakness at global optimization and the problem of its falling into a local optimum easily,a genetic algorithm with the characteristics of being self-adaptive for global optimization is used and combined with a K-means algorithm to improve the results of clustering. On this basis SGKM(senior genetic K-means) hybrid clustering algorithm is proposed which uses a cosine factor to control poor individuals in the crossover and mutation operation to make sure they are not included in the next generation.The SGKM also carries out adaptive control for crossover and mutation probabilities.Thus,it takes advantage of both efficient local search of K-means algorithms and global optimization of genetic algorithms.Experimental results show that compared with other clustering algorithms based on K-means,SGKM can achieve a smaller inner clustering distance and a greater inter clustering distance.Classification accuracy also is improved in this algorithm.The SGKM clustering algorithm can exclude poor individuals well and greatly improve the clustering effect.更多还原