Abstract: According to the time series of mobile communications traffic, this paper adopts Support Vector Regression(SVR) model which choses its hyper-parameters based on Simulated Annealing(SA) algorithms. It compares the forecasting effects of ARIMA, ANN and SVR. The influence on traffic forecasting from three hyper-parameter selection methods such as grid, GA and SA is comparatively studied. Experimental results show that the forecasting of SA-SVR is accurate and less time-consuming, it is an effective method of forecasting mobile communication traffic.

Key words: mobile communication traffic forecasting, Simulated Annealing(SA) algorithms, Support Vector Regression(SVR)

摘要： 根据移动通信话务量的时间序列，采用基于模拟退火(SA)算法对超参数选择的支持向量回归机(SVR)进行建模预测。比较ARIMA、人工神经网络和SVR 3种模型的预测效果，并对比研究网格法、遗传算法和SA 3种SVR超参数选择方法对预测效果的影响。实验结果表明，SA-SVR预测精度高、耗时少，是一种预测移动通信话务量的有效方法。

关键词: 移动通信话务量预测, 模拟退火算法, 支持向量回归机

CLC Number: 

• TP181