Research On Key Technologies Of Network Fault Diagnosis

It is important to improve the capability of the network fault diagnosis and implement a fast diagnosis scheme for keeping and enhancing the robustness of network, improving the repair ability of network and guaranteeing accomplishment of the critical missions. At the same time, enhancing capability of the network diagnosis is also necessary to reduce mean time to repair of network, and lower the maintenance cost of network. Now the fault diagnosis is hard to carry out due to many intricate correlations and uncertain factors in network. Therefore, the researchers try their best to explore the diagnosis models and methods which can describe variant network fault information accurately and deal with the uncertain knowledge effectively.This dissertation focuses on the diagnosis models and methods in order to resolve faults rapidly with low cost under the large scale network environments. After introducing the requirements of the network management, the dissertation presents the main problems that the network diagnosis faces. They are uncertain problems and diagnosis cost problems. We analyze the common diagnosis methods and focus the network diagnosis research work of this dissertation. Then we propose the object-oriented fault diagnosis models, the fault classification algorithms and the diagnosis decision algorithms. On the basis of the research work, we design and implement a network fault management system. The main contents of the dissertation are as follows:1. Aiming at the problems in the network diagnosis methods, such as the inconsistency of knowledge expression and the ambiguity of node definition, we propose a network fault diagnosis model, named SFA, which consists of symptom nodes, fault hypothesis nodes and action nodes, and then we set forth the mathematic description and knowledge characteristic of the SFA model. Based on the SFA model, we also introduce the object-oriented knowledge expression method and establish the object-oriented fault diagnosis model, named OOSFA.2. There are some shortcomings in the current classification algorithms, such as the irregularity of training samples, the lack of features selection criteria and the weakness of studying ability. According to the OOSFA diagnosis model, we bring forward two algorithms, one is the self-adaptive Bayesian classification algorithm named SFC-Bayes, which imports feature selection policy and learning policy. The other is the anti-noise support vector machines classification algorithm named AFC-SVM, which imports noise samples optimizing policy, feature selection policy and multi-level support vector machines policy. Finally, we analyze applicability of the algorithms SFC-Bayes and AFC-SVM on the condition of correlations change between the features.3. Since the current diagnosis decision methods can't work well under the actions...