生产制造企业网络故障分析与健康评估技术研究

发布时间：2018-05-07 14:10

本文选题：工业以太网 + 智能算法　；参考：《北京交通大学》2017年硕士论文

【摘要】：近些年来,随着生产制造企业网络规模的不断扩大,对网络的管理变得越来越困难。加上工业网络所处的特殊的环境,很容易造成设备故障,从而给企业带来经济损失甚至是人员伤亡。如何对网络进行有效的管理,是当前网络安全领域研究的重点。网络故障诊断与健康评估技术是根据系统当前及历史数据判断故障类型及评估系统当前健康程度的智能化技术,通过对系统进行建模及分析,可以尽快找到产生故障的原因以及从整体上把握网络的健康状况,从而给维修人员提供决策指导。本文在国家863项目的支撑下,重点研究了工业以太网络故障诊断及健康评估技术。主要研究内容如下:一是设计了综合监控管理平台,该平台负责设备状态数据采集,平台的优势是通过在管理端驻留插件的方式采集各设备的参数信息,使得对设备的管理更加灵活。二是根据生产制造企业网络的数据特点,提出了一种基于随机森林的智能故障诊断算法(CSRF)。该算法从样本采样和模型组合两方面对随机森林进行改进。前者使用分类采样技术为每个基本分类器单独生成训练样本,缓解了采样偏置和数据不均衡带来的问题。后者综合考虑了基本分类器的投票数和置信度两方面因素,提高了诊断的准确率。三是针对目前网络健康评估技术存在的问题,提出了一种多神经网络融合的健康评估算法(MNN)。该算法充分考虑到网络的单点特征以及链路特征,使用卷积网络和BP网络对不同纬度的特征进行建模分析,从而评估网络健康状况。四是通过实验设计及结果分析,验证了本文提出的算法的有效性。
[Abstract]:In recent years, with the continuous expansion of manufacturing enterprises network, network management becomes more and more difficult. Combined with the special environment of industrial network, it is easy to cause equipment failure, which brings economic losses and even casualties to enterprises. How to manage the network effectively is the focus of the research in the field of network security. The network fault diagnosis and health assessment technology is an intelligent technology to judge the fault type and assess the current health degree of the system according to the current and historical data of the system. Through the modeling and analysis of the system, the network fault diagnosis and health assessment technology is introduced. It can find out the cause of the failure and grasp the health condition of the network as a whole, so as to provide decision guidance for the maintainers. Supported by the National 863 Project, this paper focuses on the industrial Ethernet network fault diagnosis and health assessment technology. The main research contents are as follows: first, a comprehensive monitoring and management platform is designed, which is responsible for the state data collection of the equipment. The advantage of the platform is to collect the parameter information of each device by the way of staying in the management terminal. Make the management of equipment more flexible. Secondly, according to the data characteristics of manufacturing enterprise network, an intelligent fault diagnosis algorithm based on random forest is proposed. The algorithm improves the random forest from two aspects: sample sampling and model combination. The former uses classification sampling technique to generate training samples for each basic classifier separately, which alleviates the problems caused by sampling bias and data imbalance. The latter improves the accuracy of diagnosis by considering the voting number and confidence of the basic classifier. Thirdly, aiming at the problems existing in the current network health assessment technology, a multi-neural network fusion health assessment algorithm is proposed. Considering the single point feature and link feature of the network, the algorithm uses convolutional network and BP network to model and analyze the characteristics of different latitudes, so as to evaluate the health status of the network. Fourth, the effectiveness of the proposed algorithm is verified by experimental design and result analysis.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.06

【参考文献】