首页|基于深度学习的轨道交通乘客年龄属性推断方法

基于深度学习的轨道交通乘客年龄属性推断方法

扫码查看
针对轨道交通乘客年龄属性信息缺失以及难获取的问题,融合地铁AFC和城市土地利用等多源数据挖掘反映乘客年龄的出行特征,提出一种深度神经网络和自动编码器结合的乘客年龄属性推断模型.首先利用多源数据从时间和空间角度分析并提取与乘客年龄属性相关的6个出行特征(出行频率、出行时段分布、首次/末次出行时间、OD经纬度、出行时耗、目的地POI),构建乘客出行特征矩阵作为模型输入.特别的是,考虑服务能力的加权POI,以增强对目的地吸引强度刻画的准确性.分析不同年龄乘客到每个站点的出行频率,构建乘客出行稀疏矩阵,作为模型空间信息的补充输入.为学习并提取乘客出行特征间的关系和时空相关性,利用DNN对特征间关系进行捕捉;为学习乘客出行稀疏矩阵中隐含的空间关系,利用AE对稀疏矩阵压缩并进行编码和解码.最后,选取广州地铁进行案例分析,研究结果表明:与SVM,DT,MLP,AdaBoost等方法相比,DNN+AE模型的准确率分别提升了13.83%,8.01%,5.66%,4.98%,其中,老人的年龄属性推断精度最高,达到了77.51%,学生、成人的年龄属性推断精度分别达到了74.69%,68.89%.考虑服务能力的加权POI对乘客年龄属性推断结果有明显的改进.所提出的方法能够实现城市轨道交通乘客年龄属性推断,为智慧地铁运营提供支撑.
Inference Method of Passenger Age Attributes in Rail Transit Based on Deep Learning
Passenger age attributes in rail transit are missing and difficult to obtain.To address these issues,an inference model of passenger age attributes combining deep neural network(DNN)and automatic encoder(AE)was proposed by integrating multi-source data mining such as automatic fare collection(AFC)data of subways and urban land use data to reflect the travel characteristics of passengers at different ages.Firstly,multi-source data were used to analyze and extract six travel characteristics related to passenger age attributes from the perspective of time and space(travel frequency,travel time distribution,first/last trip time,OD latitude and longitude,travel time consumption,and POI of destination),and passenger travel feature matrix was constructed as the model input.In particular,a weighted POI of service capability was considered to enhance the accuracy of the characterization of destination attraction intensity.Secondly,the travel frequency of passengers at different ages to each station was analyzed,and the sparse matrix of passenger travel was constructed,which was used as the supplementary input of spatial information of the model.To learn and extract the relationship and temporal correlation between passenger travel features,DNN was used to capture the relationship between features.To learn the spatial relationship implicit in the sparse matrix of passenger travel,AE was used to compress the sparse matrix and encode and decode it.Finally,Guangzhou Metro was selected for case analysis.The research results show that compared with support vector machine(SVM),decision tree(DT),multilayer perceptron(MLP),and AdaBoost,the accuracy of the DNN+AE model is improved by 13.83%,8.01%,5.66%,and 4.98%,respectively.Among them,the inference accuracy of age attributes for the elderly is the highest,reaching 77.51%.The inference accuracy of age attributes of students and adults reaches 74.69%and 68.89%respectively.The weighted POI,which takes service capability into account,significantly improves the results of passenger age attribute inference.The proposed method can realize the age attribute inference of passengers in urban rail transit and provide support for smart subway operation.

Urban Rail TransitAge Attribute InferenceDeep Neural NetworkAutoencoderMulti-Source Data

许心越、陈培升、刘晶、张安忠、卢锦生、宋雨洁

展开 >

北京交通大学 交通运输学院,北京 100044

广州地铁集团有限公司 运营事业总部,广东 广州 510220

城市轨道交通 年龄属性推断 深度神经网络 自动编码器 多源数据

2025

铁道运输与经济
中国铁道科学研究院

铁道运输与经济

北大核心
影响因子:0.924
ISSN:1003-1421
年,卷(期):2025.47(1)