基于小波变换和ESN的P2P流量预测模型  

P2P traffic prediction model based on wavelet transform and ESN

作　　者：薛凯[1] 周亚建[1] 平源[1,2] 郭燕慧[1] 杨飞[3] 

机构地区：[1]北京邮电大学信息安全中心,北京100876 [2]许昌学院计算机科学与技术学院,河南许昌461000 [3]北京邮电大学服务科学与智能交通技术研究中心,北京100876

出　　处：《计算机工程与设计》2013年第4期1147-1152,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(60972077)

摘　　要：为了提高P2P流量预测的精度,提出一种基于小波变换和回声状态网络的流量预测模型。将原始P2P流量分解为不同尺度的高低频分量,根据不同分量的流量特性匹配不同参数的ESN模型分别预测,将多路预测结果整合输出。通过对不同P2P软件的流量预测结果表明,该模型的流量预测精度可达到98%以上,显著优于传统的ESN模型和最小二乘支持向量机模型。To make improvement on prediction accuracy of peer to peer （P2P） traffic, a novel model based on the ensemble of wavelet transformation and echo state network （ESN） is proposed. The original P2P traffic is decomposed into multiple compo- nents of either low or high frequency, then each of these components is sent to a unique ESN model with suitable parameters which match their characteristics for prediction. The final prediction result of the original P2P traffic is considered as a weighted sum of the aforementioned multiple predictions. Experimental results show that the prediction accuracy achieved by the proposed model are always more than 98% which outperform the traditional ESN and the least squares support vector machine （LS-SVM） significantly.

关 键 词：点对点 回声状态网络 小波变换 流量预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]