一种改进模糊kNN的云计算故障检测方法

Fault Detection Method for Cloud Computing Using Improved Fuzzyk Nearest Neighbor

在大规模复杂的云计算体系中,通过采集、分析系统数据可以了解系统运行的状态,从而发现并解决云计算故障问题.然而,目前基于监督学习的云计算故障检测方法忽略了噪声数据处理、训练样本的更新、未知类型故障的识别,影响了云计算故障检测的准确性.为此,本文定义了云计算故障模型并提出一种改进模糊k NN的云计算故障检测方法.该方法首先使用基于密度聚类的方法对初始云计算故障数据训练集进行预处理;其次根据模糊熵与互信息相结合的方法对云计算故障特征进行加权;然后根据故障特征权值以及分层检测改进模糊k NN,确定待检测云计算数据的近邻训练样本;最后通过基于最大隶属度的自学习确定待检测云计算数据的故障检测结果.通过实验表明本文方法对云计算故障检测是有效的.

In large-scale and complex cloud computing system, the state of system operation can be gained through collecting and analyzing system data. Then the faults of cloud computing will be found and solved. However, the current methods based on supervised learning ignore the noise data processing, the training samples updating and the identification of unknown type faults, which will affect the accuracy of cloud computing fault detection. In this paper,the cloud computing fault model is defined , and a cloud computing fault detection method improved fuzzy kNN is proposed. Firstly ,the initial training set of cloud computing fault data are preprocessed using the method based on density clustering. Secondly ,the feature of cloud computing fault is weighted according to the method of combining fuzzy entropy with mutual information. The fuzzy kNN is improved according to the fault feature weight and hierarchical detection, and the nearest neighbor sample of cloud computing data will be detected. Finally, the detection results of cloud computing data are determined by self-learning based on the maximum membership degree. Experimental results show that this method is effective in fault detection of cloud computing.