大型分布式信息系统故障检测研究  

Study on fault detection of large-scale distributed information system

作　　者：尹娟[1] 葛愿[1] 王炎[1] 徐旺[1] 陈鑫[1] 

机构地区：[1]安徽工程大学电气工程学院,安徽芜湖241000

出　　处：《安徽工程大学学报》2015年第5期45-52,共8页Journal of Anhui Polytechnic University

基　　金：国家自然科学基金资助项目(61203034);安徽省自然科学基金资助项目(1308085QF120)

摘　　要：通过引入流强度并搜索全部流强度量测之间的不变关系,将大型分布式信息系统建模成一个不变网络,其中节点代表流强度量测,边代表不变关系.当系统发生故障时,会导致不变网络中与故障点相关的边发生中断,且由于故障会在监测数据中进行传播,从而导致多条边发生中断以及多个节点出现异常,加大了系统故障检测的难度.为此,设计了gRank+算法,根据节点测量的异常水平对不变网络中的节点进行排序,从而实现系统故障的快速检测.最后使用3组综合数据集通过准确率、召回率和增益值3个指标来验证gRank+算法的有效性和优越性.By introducing flow intensity and searching invariant relationships among all flow intensity measurements,large-scale distributed information system is modeled as an invariant network,in which a node denotes a flow intensity measurement and a link indicates an invariant relationship.When a failure happens on a node in the invariant network,it will cause the links related to the node to be broken.More-over,the failure usually propagates among monitoring data,therefore more nodes will be abnormal and more links will be broken,which increases the difficulty of fault detection.This paper proposes an algo-rithm named gRank＋ to rank nodes in the invariant network according to the anomaly levels of nodes, which will realize the rapid detection and location of system failure.Finally,the effectiveness and superi-ority of the proposed method is illustrated by using three synthetic data sets in precision rate,recall rate and gain value.

关 键 词：大型分布式信息系统 不变网络 故障检测 gRank+ 算法 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]