作　　者： (何勇）; (李艳婷）;

机构地区： 上海交通大学机械与动力工程学院,上海200240

出　　处： 《工业工程与管理》 2017年第4期79-84,共6页

摘　　要： 城市移动通信基站流量的准确预测对于关键基站的拥堵控制、基站新址的选择有着重要作用。基站流量数据不仅是区域的静态表现,同时也反映区域人员的流动特性。基站流量具有非线性混沌特性,而传统的线性时间序列方法比如自回归移动平均模型难以有效地捕获实际基站流量序列中复杂的非线性因素。同时,仅考虑单个基站时间序列而忽略邻近基站的影响并不能反映基站流量的动态特征。基于向量自回归模型(VAR)对大规模基站流量数据进行整体分析,将多响应变量预测问题转化为单响应变量预测模型,运用Lasso变量选择方法筛选目标基站的重要关联基站。实例表明,相对于传统预测方法,VAR-Lasso类方法不仅提高了基站流量的预测精度,同时也实现了大规模基站的实时预测。 Traffic flow forecasting is important for the overflow control of central hubs and site selection of new communication base stations. Traffic flow data not only expresses the static attributes of specific region, but also reflects the characteristics of regional personnel mobility. Due to its nonlinear chaos property, the traditional time series model cannot capture nonlinear features of traffic flow. Some researches focused on the combination of machine learning method to improve the predictive accuracy. Considering the inner relationship between adjacent base stations,a fairly simple model known as vector auto-regression （VAR） was established for traffic flow prediction as whole. The lasso （least absolute shrinkage and selection operator） method was integrated into the model to dig out the key base stations of all. The comparison with traditional models shows that VAR-lasso is able to provide reliable forecast performance.

关 键 词： 城市基站 流量预测 向量自回归 变量选择