密 级： 学校代码：10075 分类号： 学 号：20091269 工学硕士学位论文 基于支持向量机回归的网络流量预测基于支持向量机回归的网络流量预测 学位申请人： 麻英晖 指 导 教 师 ： 李昆仑 教授 学 位 类 别 ： 工学硕士 学 科 专 业 ： 通信与信息系统 授 予 单 位 ： 河北大学 答 辩 日 期 ： 二一二年六月 Classified Index: CODE: 10075 U.D.C: NO: 20091269 A Dissertation for the Degree of M. Engineering Network Traffic Prediction Based on Support Vector Regression Candidate: Ma Yinghui Supervisor: Prof. Li Kunlun Academic Degree Applied for: Master of Engineering Specialty: Comm. it can record and reflect the activity of users. So, the prediction of network traffic could provide an effective basis for network bandwidth allocation, flow control, routing control, admission control, security management and so on. The Support Vector Machine (SVM) is a new and promising classification and regression technique proposed by V. Vapnik based on statistical learning theory. As a standard kernel learning algorithm, it has successfully been used in image processing, text classification and biological information processing, and also in time series prediction. In AdaBoost algorithm, for those samples who makes classifiers easily to make wrong points of classification are learned and vote for a second time. According to the result, the group of stronger consistency will be choosed as the foundation of the integrated study. Therefore, the repeat study of the “high error area” is positive and obvious. Inspired by Adaboost, an ensemble strategie is proposed in this paper that a further focused learning based on initial study: (1) Using LS- SVR in preliminary training: in a number of experiment and comparative experiments, the results show that the curve of flow is accurate, it is able to reflect the target movements, and training time is short, the limitations is that the distortion of details is obvious, thus, uses a single LS- SVR learner as a base is suitable, and there is space of level up. (2) Based on the fitting error of initial training by LS- SVR, we use SVM classifier to find areas (high error area) which need to be focused learning. (3) A weighted based voting algorithm is applied in the ensemble learning, the results of the study will replace to the corresponding position. The experimental results show that, the study strategy of this paper is effective, combined with the advantages of single learning and ensemble learning, it makes a balance between the speed and accuracy of training. Keywords SVM LS- SVR Ensemble learning AdaBoost Focused Learning 目 录 III 目 录 第 1 章 绪 论 . 1 1.1 课题研究的背景和意义 . 1 1.2 网络流量预测的现状 . 3 1.3 网络流量预测的可行性 . 4 1.4 本文方法 . 5 第 2 章 相关基础理论 . 7 2.1 统计学习理论基础 . 7 2.1.1 学习问题的表述 . 7 2.1.2 VC 维理论 . 7 2.1.3 推广性的界 . 8 2.1.4 结构风险最小准则 . 10 2.2 支持向量机 . 11 2.2.1 核函数 . 11 2.2.2 支持向量机分类 . 12 2.2.3 支持向量机回归 . 15 2.3 本章小结 . 17 第 3 章 集成学习 . 18 3.1 集成学习的基本方法 . 18 3.1.1 Bagging 算法理论分析 . 18 3.1.2 Boosting 算法理论分析 . 19 3.2 Adaboost 学习算法 .