基于先验知识的移动通信话务量预测

Mobile Communication Traffic Forecasting with Prior Knowledge

本文提出了一种基于先验知识引导的极大重叠离散小波变换的移动通信话务量预测方法.采用傅里叶谱分析作为小波分解子成分先验知识降低小波分解的盲目性.利用具有明确物理意义且更易提取子层的极大重叠离散小波变换对话务量序列进行分解.分解后仍以傅里叶谱先验知识为参考,合并相关子层形成趋势项和周期项两部分,并采用季节性求和自回归滑动平均(ARIMA)模型对二者分别建模和预测.采用真实数据测试的结果表明:本文方法可实现多步预测,且预测精度优于单纯的季节性ARIMA模型.

This paper proposed a methodology of forecasting for mobile communication traffic with maximal overlap discretewavelet transform （MODWT） according to priori knowledge. Fourier spectrum was chosen as the priori knowledge to avoid theblindness of wavelet decomposition. Then,MODWT which is easy to extract components with obvious physical meaning was em-ployed to decompose the communication traffic data. Moreover, prior knowledge of fourier spectrum was taken as reference to syn-thesize relevant sublayers,leading to the trend and seasonal components.Further,seasonal autoregressive integrated moving average（ARMA） model was applied to model and predict the previous trend and seasonal components,respectively. The results tested withreal communication traffic data indicate： the methodology proposed in this paper can realize multistep prediction and the forecastingaccuracy is superior to that of seasonal ARIMA models.