都市快轨交通·第30卷第2期2017年4月
doi; 10.3969/j. issn.1672 -6073.2017.02.007
基于非参数回归的
《学术探讨
城轨实时进出站客流预测
谢俏，李斌斌2，何建涛"，姚恩建2
（1.广州地铁集团有限公司，广州510030；2.北京交通大学交通运输学院，北京100044）
摘要：为准确预测城轨实时进出站客流，构建基于非参数回归的实时进出站客流预测模型。首先，对不同特征日分时进出站客流量进行对比分析，据此构建历史数据库；其次，通过计算历史分时数据的相关系数，并设置阔值对分时客流数据间的相关性进行判断，从而确定合适的非参数模型状态向量；再次，根据K近邻样本与预测目标的客流量差异性，设计基于权重加权的预测算法；最后利用广产州市城轨客流数据对预测模型进行精度分析，对全网站点多天的预测结果显示：全天平均绝对百分比误差均在2%以下，分时平均绝对百分比误差均在14%以下，表明模型具有较高的预测精度和良好的适用性。
关键词：城市轨道交通；进出站客流；实时预测；K近邻；非参数回归
中图分类号：U231
文献标志码：A
文章编号：1672-6073(2017)02-0032-05
Real-timeForecastingofEntranceandExitPassengerFlowsforUrban
RailTransitStation：ANon-parametricRegressionApproach
XIE Qiao', LI Binbin?, HE Jiantao', YAO Enjian2(1. Guangzhou Metro Group Co., Ltd., Guagnzhou 510030;
2. School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044)
Abstract: The short - term fluctuations of passenger flows should be responded quickly with the help of real - time forecasts to guarantee safe transportation. A non - parametric regression model is established to accurately forecast the real - time entrance and exit passenger flows in urban rail transit stations. Firstly, the time sharing data for entrance and exit passenger flows of dif-ferent days are compared and analyzed to lay a foundation for the construction of historical database. Secondly, the appropriate state vector for the non parametric model is defined by calculating the self - correlation coefficient of historical time - share pas-senger flow data and setting the threshold value of correlation to judge the data dependency. Thirdly, the forecasting algorithm is designed according to the entrance and exit passenger flows' difference between K -nearest neighbor samples and prediction objec-tives. Finally, the data of entrance and exit passenger flows collected from Guangzhou metro system is used for the case study, and the result shows that the mean absolute percentage errors for the day and time -sharing passenger flows are successfully limited to 2% and 14% respectively, which demonstrates that the forecasting accuracy of the proposed model is satisfactory
Keywords : urban rail transit; entrance and exit passenger flows; real -time forecast; K -nearest neighbor; non parametric re-gression
1
研究背景
随着城市轨道交通网络格局的逐步形成，网络客收稿日期：2016-09-29修回日期：2016-11-27
第一作者：谢俏，女，本科，线网管控中心副总经理，铁道工程（站场）
工程师，轨道交通运输管理方向，xieqiao@gzmtr.com
通信作者：李斌斌，男，博士研究生，交通运输规划与管理方向，
16114203@bjtu.edu.cn
基金项目：中央高校基本科研业务费专项资金资助(2016YJS066) 32URBAN RAPID RAIL TRANSIT
流规模持续攀升，地铁运营压力日益凸显。运营管理部门需要实时掌握未来短时间内客流量的变化趋势，以制定和实施合适的运营管理及客流组织计划[1]。因此，需要利用数据挖掘技术，深人剖析实时客流变化规律，滚动精准地实时预测网络客流分布状态和趋势，实现高效、精准的客流预测和预警，诱导乘客合理有序出行，节约乘客出行成本。