基于深度学习的慢行交通方式选择行为预测模型  

Prediction Model of Non-motorized Traffic Mode Selection Behaviors Based on Deep Learning

作　　者：陈文强[1] 王雪梅[1] 王涛[1] 高超 李琼[1] CHEN Wen-qiang;WANG Xue-mei;WANG Tao;GAO Chao;LI Qiong(School of Transportation Engineering,Chang an University,Xi'an Shaanxi 710064,China;Shaanxi Huidetong Municipal Engineering Co.,Ltd.,Xi'an Shaanxi 710086,China)

机构地区：[1]长安大学运输工程学院,陕西西安710064 [2]陕西汇德通市政工程有限公司,陕西西安710086

出　　处：《公路交通科技》2022年第12期204-212,共9页Journal of Highway and Transportation Research and Development

基　　金：国家重点研发计划项目(2021YFE0203600);国家自然科学基金项目(71801020);教育部人文社会科学研究一般项目(18YJC630168);陕西省自然科学基金项目(2022JQ-731);陕西省社科项目(2019D013);中央高校基本科研业务费项目(300102341622,300102342601)。

摘　　要：慢行交通是解决城市交通“最后一公里”问题的重要方式,其发展受到各国政府和学界重视。相比集计模型/非集计模型,慢行交通方式选择深度学习模型处理数据能力更强,预测精度更高,但对模型有重要影响的内生潜在变量,如态度、偏好、感知等心理因素,被置于“黑箱”而得不到合理解释。为提高慢行交通方式选择模型精度和解释能力,构建了包含个人信息、建成与自然环境、态度与认知、出行信息4维33个影响慢行交通方式选择的指标体系。通过RP调查获取有效样本931条,利用Lasso-logistic回归模型筛选影响慢行交通方式选择的显著性指标,基于筛选前后数据评估构建的神经网络模型,并与支持向量机模型进行比较,验证模型精度。结果表明:时间价值、出行距离、天气、自行车专用道、骑行技能等是影响共享单车选择的显著性因素,空气质量、交通状况、道路熟悉程度、仪表态度等是影响步行选择的显著性因素,而安全意识、环保意识等态度与认知变量对慢行交通方式选择影响较弱;经Lasso-logistic回归模型对冗余变量进行筛选后,神经网络模型的准确性得到明显提升,预测精度由81.48%提高到85.65%。对于同一组数据,深度神经网络在与支持向量机分类器的对比中表现更加突出,具有较强的预测能力和泛化能力。Non-motorized traffic is an important way to solve the problem of“the last kilometer”of urban traffic,and its development has been valued by governments and academia all over the world.Compared with aggregated/disaggregated models,the depth learning model selected by non-motorized traffic mode has stronger data processing ability and higher prediction accuracy.However,the endogenous potential variables that have important influence on the model,such as attitude,preference,perception and other psychological factors,are placed in the“black box”without reasonable explanation.To improve the model selection accuracy and interpretation ability of non-motorized traffic mode,a system consisting of 4 dimensions(personal information,built and natural environment,attitude and cognition,travel information)and 33 indicators affecting non-motorized traffic mode selection is established.931 valid samples are obtained through RP survey,the significant indicators that affect the non-motorized traffic mode selection are screened by using Lasso-logistic regression model,the data before and after screening are used to evaluate the established neural network model,which is compared with the support vector machine model to verify the accuracy of the model.The result shows that(1)Time value,travel distance,weather,dedicated bicycle lanes,and riding skills are the significant factors that affect the selection of shared bicycles.Air quality,traffic conditions,road familiarity,and appearance and attitude are the main factors affecting the selection of walking.However,attitude and cognitive variables such as safety awareness and environmental protection awareness have weak influence on the selection of non-motorized traffic mode.(2)After the redundant variables are screened by the Lasso-logistic regression model,the accuracy of the neural network model is significantly improved,and the prediction accuracy is increased from 81.48%to 85.65%.For the same set of data,the deep neural network performs more prominently in comparison with

关 键 词：城市交通 预测模型 深度学习 慢行交通 交通方式选择 

分 类 号：F570.71[经济管理—产业经济]