免疫理论在网络多重故障诊断中的应用

发布时间：2018-03-17 19:07

本文选题：多代理系统　切入点：高斯概率模型　出处：《太原理工大学》2014年硕士论文　论文类型：学位论文

【摘要】：通信技术和计算机技术的迅猛发展使网络中的故障呈现复杂化、多样化,而传统的诊断方法和现有的智能诊断技术通常只能诊断出单一的故障类型和设备,已不能满足目前网络的需求。为解决网络多重故障,可以基于单故障诊断领域研究和发展较为成熟的理论和技术,进一步研究与之相对应的智能技术,实现网络故障诊断领域追求的高效且精确的诊断系统这一主要目标。在深入理解了生物免疫系统免疫机制的基础上,进一步研究了人工免疫系统的否定选择算法,针对目前网络发展阶段的特点,提出了一种以多代理技术为主的网络故障诊断框架。该框架由两个代理组成：中央免疫代理和本地诊断代理。中央免疫代理的作用是查看、管理和做出故障决策指令等。本地诊断代理由四个不同功能的模块组成。当本地诊断代理感知到外界有诊断请求传入时,启动相应的诊断服务调度管理并协作,进行握手后网络信息采集和处理模块采集本地的数据,并进行处理作为待检测数据,然后控制、管理模块将自体与待检测数据送到故障诊断模块进行诊断,并将诊断结果传递给故障响应模块,故障响应模块进行相应记录后再将诊断结果通过调度握手后传递给中央免疫代理。本文采用BP神经网络和证据理论的合成法则实现了故障诊断模块中的多重故障诊断功能。为保证抗体多样性,提高BP神经网络的识别精度,通过高斯人工免疫系统来求解BP神经网络的权值和偏差。对BP神经网络输出节点的结果进行规范化,并作为各故障类型的基本可信度分配,根据证据合成法则计算各故障类型合成后的基本可信度分配或信度函数,最后通过判定条件实现对故障的最终判断。为获取更准确的网络状态,不占用过多的网络带宽,在实验过程中应用了动态轮询的方法采集网络状态信息。并对否定选择算法和证据理论的阈值进行了分析和研究,实验验证,本文所建框架对网络多重故障的诊断切实可行。
[Abstract]:The rapid development of communication technology and computer technology make the faults in the network is complicated, diversified and intelligent diagnosis technology of traditional diagnostic methods and existing usually can diagnose the faults and single equipment, has been unable to meet the current network demand. In order to solve the multiple fault network, single fault diagnosis theory and technology the research and development of more mature based on the further study of the intelligent technology and the corresponding, the main goal of the field of network fault diagnosis, and accurate diagnosis system.
Based on deep understanding of the immune mechanism of biological immune system, further study of the negative selection algorithm of artificial immune system, according to the characteristics of network development, presents a multi agent technology based network fault diagnosis framework. The framework consists of two components: central immune agent agent and local diagnostic agent. The central immune agent acts as a check, management and make fault decision instructions. The local diagnostic agent consists of four different modules. When the local diagnostic agent perceived external diagnostic request comes in, start the corresponding diagnostic service scheduling management and collaboration, to shake hands after network information acquisition and processing module of the local collection data, as the data to be detected and processed, and then control the management module and the detection of autologous data to diagnose the fault diagnosis module, and the diagnosis. The result is passed to the fault response module, the fault response module is recorded and then passed to the central immune agent after the handshake.
The synthesis method of BP neural network and evidence theory to realize the function of fault diagnosis of multiple fault diagnosis module. In order to ensure the diversity of antibodies, improved BP neural network recognition accuracy, weights and biases to BP neural network for solving Gauss by artificial immune system. The output of the BP neural network node results standardization. And as the basic probability assignment of the fault type, according to the evidence combination rule calculation of each fault type after the synthesis of basic probability assignment or reliability function, finally determine the conditions to achieve the final judgment of the fault.
In order to obtain more accurate network state, do not take up too much network bandwidth, in the course of the experiment methods of collecting network state information. The application of dynamic polling and the negative selection algorithm and evidence theory threshold are analyzed and experimental verification, the framework for diagnosis of multiple fault network is feasible.

【学位授予单位】：太原理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06

【共引文献】