认知网络中基于流量预测的负载均衡算法的研究

发布时间：2018-03-09 03:25

本文选题：认知网络　切入点：组合神经网络　出处：《杭州电子科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：传统网络中存在资源利用率低、业务延时明显和负载均衡效果差等问题,认知网络的出现为解决以上问题提供了可能。相比传统网络而言,认知网络具有学习推理和智能决策等特点,通过实时、动态地感知当前网络状况,进行网络智能规划、参数配置和流量调度,有效地提高网络资源利用率,并且提供更高的QoS业务。本文提出了认知网络中基于改进的组合神经网络流量预测模型WFLNN(Wavelet Fast Learning Neural Network)的负载均衡算法,重在解决网络负载不平衡问题,比如轻负载区域没有充分利用闲置资源,重负载区域发生拥堵而导致业务延时和丢包率增加等问题。首先,本文通过查阅文献深入研究了认知网络的概念及模型,同时对流量预测和负载均衡技术的研究现状进行了深入探讨。其次,本文介绍了目前常用的三种组合神经网络模型及其基本算法思路,并且选择以小波神经网络(Wavelet Neural Network,WNN)流量预测模型为基础对其进行改进。针对小波神经网络过拟合问题以及易陷入局部极小值和平点区域问题,设计了一种新的流量预测模型,即WFLNN模型。首先采用BP网络对数据样本进行去噪处理,将经过处理的数据样本实施小波变换,然后分别将小波变换得到的高频分量和低频分量作为RBF神经网络和Elman网络的输入进行训练,之后采用快速学习神经网络算法FLBP(Fast Learning Back Propagation)拟合数据。其中,FLBP算法是利用三种快速学习算法改进BP神经网络的新算法。它利用系统性的方法来检查学习过程中是否陷入局部极小值或者平点区域,然后跳过它们,从而始终保持寻找一个合适的方法来达到全局最优,进而提高预测的准确度。同时采用MATLAB对WFLNN模型和WNN模型进行比较,仿真结果表明WFLNN模型预测精度更高。最后,由于传统网络不能实时动态地感知当前的网络状态以及不能提前对服务器参数进行配置,导致网络出现延时明显等问题。因此,本文设计了一种认知网络中基于WFLNN预测模型的负载均衡算法。该算法采用认知网络的认知、推理和智能决策等机制,对网络当前状况进行感知,根据流量预测结果提前配置网络参数,并结合加权最小连接调度算法对网络流量进行调度,实现网络负载均衡。同时采用OPNET软件进行仿真,结果表明认知网络中基于WFLNN预测模型的负载均衡算法相比一般网络中未改进的算法具有更好的负载均衡效果。
[Abstract]:There are some problems in traditional network, such as low utilization of resources, obvious service delay and poor load balancing effect. The emergence of cognitive network provides the possibility to solve the above problems. Cognitive network has the characteristics of learning reasoning and intelligent decision. It can effectively improve the utilization rate of network resources by realtime and dynamically perceiving the current network situation, carrying out network intelligent planning, parameter configuration and traffic scheduling. In this paper, we propose a load balancing algorithm based on improved WFLNN(Wavelet Fast Learning Neural Network, which focuses on solving the problem of network load imbalance. For example, the light load area does not make full use of idle resources, and the heavy load area congestion leads to the increase of service delay and packet loss rate. Firstly, this paper deeply studies the concept and model of cognitive network by consulting the literature. At the same time, the research status of traffic forecasting and load balancing technology is discussed. Secondly, three kinds of combined neural network models and their basic algorithms are introduced. Based on wavelet Neural network model, a new traffic forecasting model is designed for over-fitting of wavelet neural network and easy to fall into local minimum and point region. That is, WFLNN model. Firstly, BP neural network is used to Denoise the data sample, and wavelet transform is applied to the processed data sample. Then the high-frequency and low-frequency components obtained by wavelet transform are trained as input of RBF neural network and Elman network, respectively. After that, the fast learning neural network algorithm FLBP(Fast Learning Back propagation is used to fit the data. Among them, the fast learning algorithm is a new algorithm to improve BP neural network by using three fast learning algorithms. It uses a systematic method to check whether or not it is in the process of learning. Into a local minimum or flat point region, Then skip them, and keep looking for a suitable way to achieve the global optimum, and then improve the accuracy of the prediction. At the same time, we use MATLAB to compare the WFLNN model with WNN model. The simulation results show that the prediction accuracy of WFLNN model is higher. Finally, because the traditional network can not perceive the current network state dynamically in real time and can not configure the server parameters in advance, the delay of the network is obvious. In this paper, a load balancing algorithm based on WFLNN prediction model in cognitive networks is designed, which uses cognitive network cognitive, reasoning and intelligent decision-making mechanisms to perceive the current situation of the network. According to the traffic prediction results, the network parameters are configured in advance, and the network traffic is scheduled by the weighted minimum connection scheduling algorithm. The network load balance is realized. At the same time, the simulation is carried out by using OPNET software. The results show that the load balancing algorithm based on WFLNN prediction model in cognitive network has better load balancing effect than the unimproved algorithm in general network.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN925

【相似文献】