基于小波变换和优化的SVM的网络流量预测模型  

NETWORK TRAFFIC PREDICTION MODEL BASED ON WAVELET TRANSFORM AND OPTIMISED SUPPORT VECTOR MACHINE

作　　者：周晓蕾[1] 王万良[1] 陈伟杰[1] 

机构地区：[1]浙江工业大学信息工程学院,浙江杭州310023

出　　处：《计算机应用与软件》2011年第2期34-36,59,共4页Computer Applications and Software

基　　金：国家自然科学基金(60573123)

摘　　要：提出一种基于小波变换和优化的SVM的网络流量预测模型(WaOSVM),首先对网络流量进行无抽取小波分解得到小波系数和尺度系数,然后选取适当核函数的SVM分别进行预测,其中SVM的参数用自适应量子粒子群算法(AQPSO)进行优化,最后将各预测结果进行小波重构得到最终预测结果。实验结果表明:优化过的SVM具有较好的泛化能力,该模型适合作长期预测,预测精度明显优于单一的SVM和小波神经网络模型。A new network traffic prediction model based on wavelet transform and optimised support vector machine（WsOSVM） is proposed.First,the network traffic is decomposed by non-decimated wavelet transform to acquire the scaling coefficients and wavelet coefficients,and then they are sent individually to different SVM with suitable kernel function for prediction.The parameters of SVM are optimised by adaptive quantum particle swarm optimisation（AQPSO）.At last the predictions are combined into the final result by wavelet reconstruction.Experimental results show that the optimised SVM has better generalization performance.The proposed model is suitable for long-term forecast.Compared with the single SVM and the wavelet neural networks model,it has much better prediction precision.

关 键 词：流量预测 αTrous小波变换 SVM 参数优化 量子粒子群 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]