Abstract: In order to improve predict accuracy, a combination prediction model of network traffic is proposed based on wavelet decomposition（WA-ARIMA-LSSVM）. In view of the network flow multi-scale characteristic, the network traffic time series is decomposed, and then autoregressive moving average models is used to prediction high frequency of for network traffic which low frequency is prediction by least squares support vector machine, finally high frequency and low frequency results are reconstructed, and model performance is tested by simulation experiment. The results show that compared with other network traffic prediction models, WA-ARIMA-LSSVM can accurately reflect the complex change trends of network traffic and improves the prediction accuracy of network traffic.

Key words: network traffic, Autoregressive Integrated Moving Average Model（ARIMA）, Least Squares Support Vector Machines（LSSVM）, wavelet transform, combination prediction

摘要： 为了提高网络流量的预测精度，利用小波变换、差分自回归移动平均模型和最小二乘支持向量机等优点，提出一种基于小波变换的网络流量预测模型（WA-ARIMA-LSSVM）。针对网络流量多尺度特性，首先对网络流量时间序列进行小波分解，然后分别采用差分自回归移动平均模型和最小二乘支持向量机对网络流量的高频和低频进行建模与预测，最后小波重构高频和低频的预测结果，并采用仿真实验对模型性能进行分析。结果表明，WA-ARIMA-LSSVM提高了网络流量的预测精度，可以更加准确地描述网络流量的非平稳变化趋势。

关键词: 网络流量, 差分自回归滑动平均, 最小二乘向量机, 小波变换, 组合预测