【作者】 李千目；

【导师】 刘凤玉；

【作者基本信息】 南京理工大学 ， 计算机应用技术， 2005， 博士

【摘要】 随着新一代战略互联网规模的不断扩大,网络应用不断增加,传统的网络故障诊断系统功能单一、操作复杂、效率低下,已不能满足军网管理的发展需要。如何有效地、安全地、易扩展地管理网络故障是目前迫切需要解决的问题,其中,故障检测、故障定位是故障管理中的关键环节。本文由网络故障的层次传播性出发,从信息的可用性角度,构建一个战略互联网故障诊断问题的合理解决方案。 本文首先针对战略互联网的管理由集中式向分布式的发展趋势,依据动态SNMP代理群的思想,讨论了动态群管理策略(包括组群策略和故障管理模式变更策略),建立了一个自适应分布式诊断模型,并提出了基于易损链路的稳态群首选举算法。在此基础上,本文提出分层分散故障诊断策略,对战略互联网四层结构(物理接入层,链路传送层,网络控制层和应用层)的不同故障特点和状态属性融入不同的检测策略,给出一个较为完整的解决方案。 本文主要研究工作和取得的成果如下: (1) 基于仿生学的免疫原理,将肽链定义为网络中执行的事件检测序列,提出并实现了一种新型的物理层故障节点定位方法—基于生物免疫学的故障定位。该方法首先依据“阳性选择”原则进行事件库设计,对高频度行为模式优先分析和处理,提高了检测的速度和效率。其次,依据故障之间的事件检测序列关联关系,运用图论和邻接矩阵的方法求出根故障集,由故障相关性确定故障源,有效地起到故障过滤和定位的功能。经实验证明,本方法具有很强的实效性。 (2) 基于粗糙集的神经网络理论,提出链路传输层故障诊断的RSNN算法,实现不一致情况下的故障规则获取和学习样本的净化处理。该算法具有简化样本、适应性强、容错性高等特点,能有效处理链路传送层故障诊断中噪声和不相容的信息。由于诊断问题的实质是一种映射,该算法用一种前馈型网络来逼近这种映射关系,实现对故障的有效分类。实验表明,利用该方法实现的系统与同类的其他方法相比,大大提高了诊断准确率和诊断速度。 (3) 运用弱T范数簇模糊神经元,设计出一种基于粗糙模糊神经网络的网络控制层拥塞预测算法(RFNN)。RFNN不仅具有单调性和连续性,而且能满足网络拥塞的推理一致性要求。实验表明,利用RFNN的处理不确定性问题和自学习能力进行流量预测,与传统拥塞预测方法相比,具有较好的效果。 (4) 提出了基于SVM的网络应用层故障检测模型,并对模型各个组件的功能、机制和实现进行了深入探讨。对用于检测的网络数据特征,本文利用异构数据集上的更多还原

【Abstract】 The new generation strategic internet is a new type military network, constituted by physical layer, link transport layer, network control layer and application layer. It can flexibly support multi-service, and has anti-destroy, re-combined properties. With the development of computer science and communication network, the scale of strategic internet is growing larger, together with the emergence of more network applications. Owing to simple function, complex operation and lower efficiency, the old network troubleshooting system already can’t satisfy for the demands of carrier development. In order to perform high efficiency and reliability, it is very important for us to set up a perfect network troubleshooting system. With the development of the present distributed network troubleshooting management, a self-adapting distributed network management framework based on dynamic SNMP agent groups is presented in this paper. The explanation of how to achieve this model is discussed. Especially, dynamic groups’ management policy in the model is discussed, including grouping policy and choice of management mode, and a stable group Ω leader election algorithm based on loss links is given. According to the feature of strategic internet architecture, this paper brings layering-decentralization intelligence into this field, which makes it possible for the failure automatic location and diagnosis.The main achievements of this paper are as follows:(1) According to the immunology principles of bionics, a new physical layer nodefault location method-Immunology based Fault Location Method is presented. Inthis paper event detection sequences are viewed as analogous to peptide. According to the principle of positive selection in Immunology, the system builds up its event database. The behavior model whose frequency is higher will be analyzed and processed first. It improves the speed and effectiveness of intrusion detection. Fault Location is based on the event detection sequences correlation, graph theory and adjacency matrix are two methods to get the root failure sets. With the relationship of failures, this paper gives a method to determine the source of failure in this paper, which will perform failure filtration and location function effectively. The experiment system implemented by this method shows a good diagnostic ability.(2) Puts forward RSNN algorithm, a designing fault diagnosis method for link transport layer, which tightly combines neural network and rough sets. We can get reducedinformation table, which implies that the number of evaluation criteria is reduced with no information loss through rough set approach. And then, this reduced information is used to develop classification rules and train neural network to infer appropriate parameter. The rules developed by RS-Neural network analysis show the best prediction accuracy, if a case does match any of the rules. It’s capable of overcoming several shortcomings in existing diagnosis systems, such as a dilemma between stability and redundancy. Since the essence of fault diagnosis is a kind of mapping, an artificial neural network model is adopted to deal with the mapping relation, categorizing the network faults. The experiment system implemented by this method shows a good diagnostic ability.(3) Consisting of weak T-norm cluster fuzzy neuron, a rough fuzzy neural network (RFNN) is constructed in this paper, which is applied to network control layercongestion inference. RFNN overcomes a few shortcoming of the conventional CRI, and it is much easier to satisfy consistency principle of fuzzy inference than CRI. Analyzed the properties of the new method, we discovered that it is continuous and monotonic. The reasoning results prove better performance obtained than other conventional congestion methods.(4) A framework of SVM based Network Fault Detection System of Application Layer is proposed. The function, mechanism and realization of the components of this framework are discussed in the paper. By means of distance metric of heterogeneous datasets, the feature data of network are preprocessed. Based on guaranteed estimators, we estimate the size of test set. Thus we not only avoid bad train result for lack of examples, but also reduce the training time and improve the efficiency of training. During the training, by means of fuzzy mathematics, considering the effect of different network data features to the classification, a weight method is brought forward. It improves the accuracy of network fault detection. The problem of low detection accuracy of some types of faults for the imbalance of training examples is researched. A method of increasing the proportion of the examples of these types is presented. It improves the detection accuracy of these types of faults.(5) A framework model proposed in this paper is a data-link redundant strategy based on reliability theory. The redundant running(RUS) could be combined with the normal maintenance, which greatly improves the performance of the network system. The static-checking and policy of authentication mechanism ensure the running network without any error. The redundant equipments are independent but are capable of communication with each other when they work their actions. The model is independent of更多还原