基于MAWILab数据集的短期网络流量预测

Short-term Prediction for Network Traffic based on MAWI Lab Dataset

针对网络性能优化中短期流量预测模型选择问题,设计仿真,采用支持向量回归、神经网络(多层感知器)和线性时间序列(ARIMA)等多种模型,对MAWILab数据集中骨干网流量进行短期预测。应用一步预测方法,得到不同流量序列样本的预测结果,通过量化预测误差比较不同模型的预测性能,得出最终结论,在实际网络流量短期预测问题中,与ARIMA模型相比,具有非线性函数拟合功能的支持向量回归和神经网络可以取得更好的预测精度。

For choosing a right model to predict short-term traffic in network performance optimization tasks,simulations wese designed to use support vector regression,neural networks(multilayer perceptrons) and linear time series(ARIMA) models to conduct prediction of actual backbone traffic from the MAWILab dataset.One step method was applied to obtain prediction results of quite a lot actual traffic samples,and prediction errors wese compared quantitatively to validate different model’s performance.The final conclusion is that,in comparison with traditional ARIMA model,support vector regression and neural networks can achieve better prediction accuracy in short-term traffic prediction tasks.