基于蚁群算法优化的小波神经网络流量预测模型研究

发布时间：2018-04-09 12:23

本文选题：网络流量预测　切入点：小波分析　出处：《西安电子科技大学》2014年硕士论文

【摘要】：随着互联网的使用量逐渐增加,网络种类逐渐繁多,信息数据越来越庞大并且呈现出多变的动态特性,如何精确地对网络流量进行预测从而实现对网络运行状态的有效管理,成为一个日益重要的问题。其中,如何建立一个精确而高效的流量预测模型是解决该问题的重点和难点,其对于网络的测试、流量的控制以及网络的设计规划和质量保证等问题的研究,都有重要的意义。本文主要对网络流量预测模型进行研究和设计,具体工作有三个方面:(1)介绍了网络流量预测的起源、发展状况及其研究的意义,同时对网络流量所具有的特性进行了分析和描述;(2)重点介绍了小波分解理论和BP神经网络的工作原理,并对两者组合而成的小波神经网络(Wavelet Neural Network,WNN)进行了介绍,同时着重分析了基于GA算法的WNN预测模型;(3)针对基于GA算法优化的WNN预测模型存在的不足,本文提出用具有正反馈机制和启发式学习特点的蚁群算法(Ant Colony Optimization,ACO)代替GA算法,形成一种基于ACO算法优化的WNN预测模型—WACONN预测模型。该模型为了克服BP算法在网络训练过程中对初始权值、阈值比较敏感,且容易陷入局部最优解的缺点,提出将ACO算法与BP算法相结合形成ACO-BP算法:首先利用ACO算法的全局寻优能力为BP神经网络提供较适合的初始权值和阈值,以降低BP算法由于对初始值设置不同而产生较大的差异性、且容易陷于最优解等缺点,再利用BP算法进一步迭代寻取全局最优解。在进行仿真实验过程中,该模型首先利用同时具有时、频域描述特性的小波变换,将数据样本分解成高频分量和低频分量;然后利用高、低频分量的训练集部分对基于ACO优化的WNN预测模型进行训练;接下来,将高、低频分量的预测集部分输入训练好的模型,进行预测;最后将预测得出的高、低频数据分量相叠加,得出最终的预测结果。实验数据表明,与基于GA算法优化的WNN预测模型相比,本文提出的基于ACO算法优化的WNN预测模型具有更好的预测精度和更快的网络收敛速度,是一种更有效的预测模型。
[Abstract]:With the increasing use of the Internet, the variety of the network is becoming more and more numerous, the information data is becoming more and more large and showing dynamic characteristics, how to accurately predict the network traffic so as to realize the effective management of the network running state.Become an increasingly important issue.Among them, how to establish an accurate and efficient traffic prediction model is the key and difficult point to solve this problem. It is of great significance to the research of network testing, traffic control, network design, planning and quality assurance.This paper mainly studies and designs the network traffic forecasting model, and introduces the origin, development and significance of network traffic forecasting in three aspects.At the same time, the characteristics of network traffic are analyzed and described. (2) the wavelet decomposition theory and the working principle of BP neural network are emphatically introduced, and the wavelet Neural network WNN which is composed of the two neural networks is introduced.At the same time, the WNN prediction model based on GA algorithm is analyzed emphatically. Aiming at the shortcomings of WNN prediction model optimized by GA algorithm, this paper proposes to replace GA algorithm with ant colony algorithm with positive feedback mechanism and heuristic learning characteristics.A WNN prediction model-WACONN model based on ACO algorithm optimization was developed.In order to overcome the shortcomings of BP algorithm which is sensitive to initial weights and easy to fall into local optimal solution in the course of network training.A ACO-BP algorithm is proposed by combining ACO algorithm with BP algorithm. Firstly, the global optimization ability of ACO algorithm is used to provide a suitable initial weight and threshold for BP neural network.In order to reduce the disadvantages of BP algorithm because of the difference of initial value and easy to be trapped in the optimal solution, BP algorithm is used to find the global optimal solution.In the course of simulation experiment, the model firstly decomposes the data sample into high frequency component and low frequency component by wavelet transform with simultaneous time and frequency domain description characteristics, and then uses high frequency component.The training set of the low frequency component trains the WNN prediction model based on ACO optimization. Then, the high and low frequency component prediction sets are input into the trained model to forecast.The low frequency data components are superposed and the final prediction results are obtained.Experimental data show that compared with the WNN prediction model optimized by GA algorithm, the proposed WNN prediction model based on ACO algorithm has better prediction accuracy and faster network convergence speed, so it is a more effective prediction model.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06;TP18

【相似文献】