基于LightGBM和SHAP的地铁短时进站客流预测及影响因素分析  

Metro Short-term Passenger Flow Forecasting and Influence Factors Analysis Based on LightGBM and SHAP

作　　者：聂小虎[1] 吴东平[1] 安静仪 常红光 阳旭明 NIE Xiaohu;WU Dongping;AN Jingyi;CHANG Hongguang;YANG Xuming(China Communications Construction Second Highway Consultants Co.,Ltd.,Senior Engineer,Wuhan 430052,China;Wuhan University of Technology,Wuhan 430063,China)

机构地区：[1]中交第二公路勘察设计研究院有限公司,湖北武汉430052 [2]武汉理工大学交通与物流工程学院,湖北武汉430063

出　　处：《铁道经济研究》2023年第5期50-55,共6页Railway Economics Research

摘　　要：为更精确地预测地铁短时进站客流量,本文提出一种基于LightGBM的轨道交通短时客流预测方法,以15 min为粒度计算杭州市各地铁车站进站客流并作为因变量,提取时间、空间、车站属性、气象要素和土地开发五个维度的自变量,构建模型训练数据集。通过贝叶斯搜索优化模型参数,选择六项评价指标与常用模型进行对比,并利用SHAP方法解析各影响因素的重要度及与预测结果的关系。结果表明:Light‐GBM模型预测误差显著较小,对地铁短时客流预测能力优秀;影响短时客流大小排名前4的因素依次为:前一时间段客流量、前一周同时段客流量、前一天同时段客流量和前3天同时段客流均值,其中排名第一的因素对短时客流变化具有主导作用;而后三个因素取值分别达到2000、1500和500人次的阈值后,对客流量的增进作用不再显著变化。To more accurately predict the short-term inbound passenger flow of the metro,we propose a short-term passenger flow prediction method for the metro based on LightGBM,using 15min as the granularity to calculate the inbound passenger flow of all metro stations in Hangzhou and as the dependent variable,and extracting the independent variables in five dimensions:time,space,station attributes,meteorological elements and land development,to construct a model training data set.By optimizing the model parameters through Bayesian search,we select six evaluation indicators for comparison with commonly used models,and use the SHAP method to analyze the importance of each influencing factor and its relationship with the prediction results.The results show that the prediction error of the LightGBM model is significantly small,and its ability to predict metro short-term passenger flow is excellent.The top 4 factors affecting the ranking of short-term passenger flow are passenger flow in the previous period,passenger flow in the same period in the previous week,passenger flow in the same period in the previous day,and average passenger flow in the same period in the previous 3 days.Among them,the top-ranked factor has a dominant role in short-term passenger flow changes.After the values of the latter three factors reach the thresholds of 2000,1500,and 500 passengers respectively,the enhancement effect on passenger flow no longer changes significantly.

关 键 词：交通工程 短时客流预测 影响因素分析 LightGBM SHAP 轨道交通 

分 类 号：U293.6[交通运输工程—交通运输规划与管理]