论文总字数：19849字

摘 要

如今网络发展日益迅速、网络服务日益增多、网络环境日益复杂，因此网络流量预测技术变得尤为重要。面对海量的网络流量数据，网络流量预测需要经得起多重考验，在保证精准度的同时还需要能适应各类网络环境，因此一些传统的网络流量预测技术已经不太适用了。

随着深度学习的发展，人们将深度学习应用于各个领域，网络流量预测就是其中一个领域。卷积神经网络是深度学习模型中的佼佼者，所以基于卷积神经网络的网络流量预测算法是近年来兴起的一种新的网络流量预测方法。卷积神经网络对流量预测采用的方法是将网络流量转换成图像信息，图像信息经过特有的卷积层和池化层结构映射到预测模型内部。由于卷积神经网络的权值共享性质，基于卷积神经网络的网络流量预测工作比其他神经网络预测方法简便。

针对传统网络流量预测模型预测精度低，不适用于大量网络历史流量的问题，本文提出了基于卷积神经网络的网络流量预测算法。算法通过设置最佳的卷积核尺度、池化方式、优化器和神经网络深度来寻求最优的网络流量预测模型，以获得最优的流量预测精度。

关键字：网络流量预测 深度学习 卷积神经网络 优化器

Traffic Prediction Algorithm Based on Convolutional Neural Network

Abstract

With the rapid development and popularization of the network, network services are increasing day by day, and the network environment is becoming more and more complex. The prediction technology of network traffic has become particularly important. Facing with massive amounts of network traffic data, the network traffic forecasting needs to withstand multiple tests. It also needs to adapt to various network environments with ensuring accuracy. Therefore, some traditional techniques of network traffic forecasting are no longer applicable.

With the development of deep learning, people apply deep learning to various fields, and the network traffic prediction is one of them. The convolution neural network is a leader in deep learning models, so the network traffic prediction algorithm based on convolution neural network is a new network traffic prediction method that has arisen in recent years. The convolution neural network converts network traffic into image information, which is mapped into the prediction model through the special convolution layer and pooling layer structure. Due to the weight sharing property of convolutional neural networks, the network traffic prediction based on convolutional neural networks is simpler than other neural network prediction methods.

The traditional network traffic prediction models with low prediction accuracy is not suitable for a large number of historical network traffic. In this paper, a traffic prediction algorithm based on convolutional neural network is proposed to find the best prediction precision of network traffic by setting convolution kernel size, pooling method and optimizer.

Key words: network traffic forecast, deep learning, CNN, optimizer

目 录

摘 要 Ⅰ

第一章 绪论 1

1.1 研究背景 1

1.2 网络流量预测研究现状 1

1.2.1传统网络流量预测模型 1

1.2.2深度学习预测模型 2

1.3 课题研究内容 3

1.4 论文组织结构 3

第二章 卷积神经网络概述 5

2.1 卷积神经网络的背景 5

2.1.1 人工神经网络 5

2.1.2 神经元模型 5

2.2 卷积神经网络的原理 6

2.2.1 局部连接和权值共享 6

2.2.2 卷积层和子采样层 7

2.2.3 分类器 8

2.2.4 激活函数 9

2.2.5 损失函数 9

2.3 卷积神经网络参数初始化 10

2.4 卷积神经网络的训练方法 10

2.4.1 卷积神经网络的有监督学习 10

2.4.2 卷积神经网络的无监督学习 11

2.5 本章小结 11

第三章 基于卷积神经网络的网络流量预测算法 12

3.1 网络流量数据的预处理 12

3.2 参数设置 12

3.3 仿真实验及结果分析 13

3.3.1最佳卷积核尺度选择 13

3.3.2 最佳优化器选择 13

3.3.3最佳池化方式选择 14

3.3.4 最佳网络深度选择 15

3.3.5 实验结果 16

3.4 本章小结 18

第四章 总结与展望 19

4.1 总结 19

4.2 展望 19

参考文献 20

致 谢 23

请支付后下载全文，论文总字数：19849字