一种动态加权模糊聚类算法的研究  

Dynamic weighted fuzzy clustering algorithm

作　　者：刘锐 张宁[1] LIU Rui;ZHANG Ning(School of Computer and Information Technology, Beijing Jiaotong University, Beijing 100044, China)

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044

出　　处：《铁路计算机应用》2018年第5期5-8,17,共5页Railway Computer Application

基　　金：科技部科技支撑项目(2015BAF08B02)

摘　　要：针对模糊C均值聚类(FCM)算法选取初始中心具有随机性这一缺陷,利用遗传算法优化FCM算法,根据适应度函数动态确定交叉、变异算子,从而选取最优初始中心,避免FCM算法陷入局部极小;针对FCM受噪声点、孤立点影响较大的缺陷,利用LOF加权降低数据噪声点对聚类的影响,并将FCM聚类、遗传算法、加权策略相结合,提出一种新的动态加权模糊聚类算法。经UCI通用数据集验证,优化后的聚类算法可以有效提高聚类质量和准确度。Aimed at the defect that the initial center selected by fuzzy C-means clustering （FCM） algorithm is random,this paper presented the use of genetic algorithm to optimize FCM algorithm. According to the fitness function toadaptively determine the crossover, mutation operator, thus choose the optimal initial center, avoide the FCM algorithminto a local minimum; Aiming at the defect that FCM is influenced by noise point and isolated point, this paper usedLOF weighting to reduce the impact of noise points on the clustering. Combining with FCM clustering, geneticalgorithm and weighted strategy, a new dynamic weighted fuzzy clustering algorithm was proposed. The UCI universaldata set verified that the optimized algorithm effectively improved the quality and accuracy of clustering.

关 键 词：FCM算法 遗传算法 LOF算法 权重 

分 类 号：U266.2[机械工程—车辆工程] TP39[交通运输工程—载运工具运用工程]