基于CRO高阶神经网络的流量矩阵估计研究

发布时间：2018-06-24 01:51

本文选题：流量矩阵估计 + 高阶神经网络　；参考：《华中师范大学》2015年硕士论文

【摘要】：随着互联网技术的飞速发展,生活水平的不断提高,越来越多的用户加入到互联网中,网络规模日益扩大；各种网络应用也是层出不穷,网络中传输的流量以几何式增长。对网络运营商和网络管理者而言,节约成本,优化网络规划设计,提高网络服务质量,全面掌握网络运行状态等网络工程问题亟需得到进一步解决。了解网络内部的种种特性有助于成功设计、控制和管理网络。网络流量矩阵作为重要的工具之一,全面地描述了网络上所有节点间的流量分布,是网络设计、管理和路由配置的重要依据。而庞大的网络规模,海量的传输数据,异构分布的网络模式使得直接通过网络测量获取流量矩阵非常困难,甚至不可能实现。因而广大学者提出了各种利用有限测量信息进行网络流量矩阵估计的间接测量方法。本文围绕流量矩阵估计这一课题所做的工作主要包括：1)首先介绍了流量矩阵的研究意义和目前国内外的研究现状,概要描述了本文的层次结构；2)归纳了流量矩阵间接测量的三类方法,对每一类方法中应用的技术作了基本的描述,分析了各类方法的优缺点；3)在高阶神经网络的基础上,应用最新的化学反应优化算法,提出了一种新型的CRO-PSNN算法来对流量矩阵进行估计。对于该算法,本文从两个方面论述了其优越性：一是高阶神经网络对于流量矩阵估计这类高维度病态问题的优势,包含乘法器使得高阶神经网络能够处理普通神经网络所不及的高阶问题和强非线性问题。二是使用化学反应优化算法参与高阶神经网络的学习过程,避免了以往误差参与的权值调整,减小了计算量,加快网络的收敛速度和计算速度。4)最后根据现有的理论和实验依据,通过仿真实验与知名的流量矩阵估计方法进行对比,证明该方法所具有的优势。最后总结全文,回顾了本文所做的研究工作,并根据目前的现状对进一步的研究方向作出了展望。
[Abstract]:With the rapid development of Internet technology and the improvement of living standard, more and more users join the Internet, and the scale of network is expanding day by day. For network operators and network managers, such network engineering problems as cost saving, optimization of network planning and design, improvement of network service quality and overall grasp of network operation state need to be solved further. Understanding the internal features of the network helps to successfully design, control, and manage the network. As one of the important tools, the network traffic matrix describes the traffic distribution among all nodes in the network. It is an important basis for network design, management and routing configuration. However, because of the huge network scale, the massive data transmission and the heterogeneous distributed network mode, it is very difficult or even impossible to obtain the traffic matrix directly through the network measurement. Therefore, a variety of indirect measurement methods using finite measurement information to estimate network traffic matrix have been proposed. The main work of this paper includes: 1) firstly, the research significance of traffic matrix and the current research situation at home and abroad are introduced, and the hierarchical structure of this paper is briefly described. 2) three kinds of methods for indirect measurement of flow matrix are summarized, the basic description of the techniques used in each method is given, and the advantages and disadvantages of each method are analyzed. 3) on the basis of high-order neural networks, the latest chemical reaction optimization algorithms are applied. A new CRO-PSNN algorithm is proposed to estimate the flow matrix. For this algorithm, this paper discusses its advantages from two aspects: one is the advantage of high-order neural networks for high-dimensional ill-conditioned problems such as traffic matrix estimation. The inclusion multiplier enables higher order neural networks to deal with higher order problems and strongly nonlinear problems that are beyond the reach of ordinary neural networks. The second is to use chemical reaction optimization algorithm to participate in the learning process of high-order neural network, avoid the weight adjustment of the previous error participation, and reduce the calculation amount. Finally, according to the existing theoretical and experimental basis, the simulation experiment is compared with the well-known flow matrix estimation method, and the advantages of the method are proved. Finally, the paper summarizes the full text, reviews the research work done in this paper, and forecasts the future research direction according to the present situation.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TP393.06;TP183

【参考文献】