关键词: 故障特征, 距离度量算法, 特征加权

Abstract: In order to avoid security threat in the various networks, the unite network management platform which is composed of many heterogeneous security devices has been application widely. But how to deal with lots of conflicts and faults that arise from security devices has been key point and difficulty of network management. In this paper, distance metric and feature weighted algorithms have been provided for pretreatment of data which are gathered in the network. Because fault features of network security device have quite heterogeneous character. So the twenty-three feature variables have been partitioned four classes which contain continuous attribute, integer attribute, nominal attribute and dimidiate attribute. Then Euclid, VDM, DVDM and IVDM algorithms have been used for represent differ samples’ distance. Finally, fuzzy theory and knowledge rules have been applied to feature weighted of four algorithms. In the experiment, the fault classification accuracy of the weighted algorithm of DVDM and IVDM increase more than 10 percent. It shows this theory has upper reliability and precision.

Key words: fault feature, distance metric algorithm, feature weighted