摘要
话务预测是整个通信保障工作的基础,其预测精度决定了整个规划的合理性和科学性。而节假日话务量,具有历史样本量较小和非线性强的特点,传统的预测方法很难实现精确的预测。支持向量机在解决小样本和非线性问题时表现出许多特有的优势。提出了一种改进的网格搜索法和交叉验证法对支持向量回归机(SVR)参数优化选择,并对节假日忙时话务进行预测,并与BP神经网络、基本的SVR和网格搜索SVR三种预测模型进行比较。而且用免疫算法和粒子群算法优化SVR参数与本文算法作比较来预测普通日子的话务量。实验结果表明,基于改进的网格搜索SVR预测精度高、耗时少、稳定性强,具有很好的实用性和推广性。
The traffic prediction is the basis of the whole communication＇s security work, whose prediction accuracy determines the rationality and scientificity of the entire plan. While the prediction of holiday＇s traffic has the characteristics of small historical sample size and strong nonlinear, it is hard to realize accurate prediction for the traditional prediction method. An improved grid search method for selecting the optimized parameter of Support Vector Regression machine （SVR） and then predicting the busy traffic in holidays is proposed and compared with BP neural network, SVR and grid search SVR. And the traffic of general days is predicted by comparing our method with Immune algorithm and Parti- cle Swarm Optimization algorithm in optimizing SVR parameters. The experimental results show that the improved grid search SVR has a higher forecast precision, a less time-consuming and a strong stability, thus having good practicality and promotion.
出处
《计算机工程与科学》
CSCD
北大核心
2014年第4期707-712,共6页
Computer Engineering & Science
基金
中国移动通信集团新疆有限公司发展基金项目(XJM2011-11)
关键词
节假日话务预测
支持向量回归机
改进的网格搜索法
prediction of holiday＇s traffic
support vector regression machine
improved grid searchmethod
