基于迭代式MapReduce的FCM算法实现  

Implementation of FCM Algorithm Based on Iterative MapReduce Model

作　　者：赵虎 左开伟 

机构地区：[1]武警工程大学信息工程系,西安710086

出　　处：《计算机测量与控制》2016年第11期240-242,252,共4页Computer Measurement &Control

基　　金：国家自然科学基金(61402529);武警工程大学基础研究基金(WJY201603)

摘　　要：针对模糊C均值聚类（Fuzzy C--Means Clustering，FCM）算法聚类过程迭代的特点，采用迭代式MapReduce模型对FCM算法进行了优化实现；Map函数计算每个样本到聚类中心的隶属度，Reduce函数接收Map函数的中间输出计算新的聚类中心，传递模块将最新聚类中心传送给原Map任务所在节点，供新一轮MapReducejob使用；迭代式MapReduee模型在MapReduce基本模型上添加了传递模块，有效解决了基本模型在处理迭代问题上存在的不足；在Hadoop平台中，分别使用基于迭代式MapReduce和MapReduee基本模型的FCM算法对变压器进行故障诊断；实验结果表明，基于迭代式MapReduee的FCM算法诊断速度达到了基于MapReduce基本模型算法诊断速度的12倍以上，误判率降低了12％～15％，有效提升了FCM算法的诊断效率。The FCM algorithm is iterative. How to implement it using iteration MapReduce model was studied. Map function calculate the membership degree of each sample to the cluster center. Reduce function calculate the new cluster center based on the Map results. The transmitting module transmit the new cluster center to the former Map nodes for the next MapReduce job. To solve the shortage in iteration program, the iteration model added a transmitting module. The experiment diagnosed the transformer samples on Hadoop based on the basic model and the iteration model. The diagnosing speed of FCM algorithm based on the iteration model is more than 12 times faster than the basic model. The error rate reduce 12%-15% . The iteration model effectively improve the diagnostic efficiency of the FCM algorithm.

关 键 词：MAPREDUCE FCM算法 迭代 云计算 变压器 

分 类 号：TP393.0[自动化与计算机技术—计算机应用技术]