论文部分内容阅读
为提高网络流量时间序列的中期预测精度,提出一种高斯过程回归模型补偿自回归积分滑动平均(auto regressive integrated moving average,ARIMA)模型的网络流量预测模型.首先通过BDS(Brock-Dechert-Scheinkman)统计量检验方法确定网络流量时间序列包含线性特征与非线性特征.然后利用ARIMA模型对网络流量时间序列进行非平稳建模,得到符合网络流量序列线性变化规律的模型.通过人工蜂群算法优化的高斯过程回归模型对具有非线性特性的预测误差序列进行建模与预测.最后将ARIMA模型的预测值与高斯过程回归模型的预测误差值进行相加得到最终的网络流量预测值.仿真对比实验表明提出的预测方法具有更高的预测精度与更小的预测误差.
In order to improve the medium-term prediction accuracy of network traffic time series, this paper proposes a network traffic prediction model of Gauss process regression model to compensate the auto-regressive integrated moving average (ARIMA) model.First, through the BDR (Brock-Dechert-Scheinkman) The statistical test method determines that the network traffic time series contains linear features and non-linear features, and then uses ARIMA model to model the non-stationary network traffic time series to get the model which is in line with the linear variation of the network traffic sequence.Optimized by artificial bee colony algorithm Gaussian process regression model is used to model and forecast the prediction error sequence with nonlinear characteristics.At last, the prediction value of the ARIMA model and the prediction error of the Gaussian process regression model are added to get the final network traffic prediction value.Comparison experiment shows The proposed prediction method has higher prediction accuracy and smaller prediction error.