Abstract: Prediction of network traffic flow is the foundation of network management, network traffic is affectd by lost of factors, has time-varying and nonlinear, the traditional methods are difficult to accurately describe the network flow changes, and this paper puts forward a network traffic prediction model（ARIMA-LSSVM）. The ARIMA of network traffic periodic trend is predicted by ARIMA model, then nonlinear trend is predicted by LSSVM model and LSSVM parameters are optimized by genetic algorithm, the prediction results are used as LSSVM inputs to get combined model prediction results. The simulation results show that, compared with the single network traffic prediction model, ARIMA-LSSVM improves network traffic prediction accuracy, it can more completely characterize network traffic change trend.

Key words: network traffic, Autoregressive Integrating Moving Average（ARIMA）, Least Square Support Vector Machines（LSSVM）, hybrid model

摘要： 网络流量预测是网络管理的基础，网络流量受到多种因素影响，具有周期性、时变性和非线性，传统单一线性模型ARIMA或非线性模型SVM均难以准确描述网络流量复杂变化规律，为此，提出一种网络流量组合预测模型（ARIMA-LSSVM）。采用ARIMA对网络流量进行预测，捕捉其周期性变化趋势，采用LSSVM对网络流量非线性变化趋势进行预测，同时采用遗传算法对LSSVM参数进行优化，采用LSSVM两种预测结果进行融合，得到网络流量的最终预测结果。仿真实验结果表明，相对于单一网络流量预测模型，ARIMA-LSSVM提高网络流量预测精度，更能全面刻画网络流量变化趋势。

关键词: 网络流量, 差分自回归滑动平均模型, 最小支持向量机, 组合模型