基于潜在类别模型的通勤者交通方式选择异质性研究

发布时间：2018-03-25 05:20

本文选题：通勤出行　切入点：通勤方式选择　出处：《南京大学》2017年硕士论文

【摘要】：通勤出行是城市居民最主要的出行,同时也是最重要的出行,它对城市的日常交通状况具有决定性的影响,城市交通拥堵表现最显著的是通勤时段拥堵。受限于城市交通资源,扩大道路供给、提升交通服务质量不能作为永久性措施。从通勤者个体选择行为自身特点和偏好出发,另辟新法,是近年来交通出行行为研究领域的新思路。离散选择模型是研究通勤者路径选择和方式选择的经典方式,它的发展历程也随着人们的需求而不断被改进,其中混合Logit模型是可以根据个体的随机喜好因素来预测其选择方式。而本文重点关注的潜在类别模型则是对不可测因素的进一步研究,而且它还在很大程度上做到控制变量,剔除相互干扰因素,用潜在变量衡量需考察的外显变量之间的联系,在对一个群体做出新的分类的同时,也能探求得这个群体在人群中的比重大小。有利于我们在进一步研究课题、引导出行者的时候有的放矢。潜在类别模型方法是对通勤出行者出行行为选择异质性研究的重要补充。本文具体的研究内容和主要结论如下:首先,通过对通勤出行的重新认识,了解城市通勤出行的空间特点、强度特点、目的特点。其次,基于传统模型的出行行为选择方式研究领域的成果,认识到在很多情形之下,增修道路、高架和隧道等方法只是短期内有明显效果。接下来介绍并分析潜在类别模型的数学原理,通过几个基本算例来区分与既有分类的方式细分人群的差别。最后,通过陈述性调查法收集数据,运用潜在类别模型对调查人群进行分类,得到每项属性的参数估计和影响因子系数,并给每个类别命名,分析其类别特点。本文通过对经济水平、天气状况、通勤距离、通勤花费、出勤紧要、舒适度、身份认同感这七项属性被选择来做陈述性调查,每项属性选择的分为三个水平,通过LatentGold软件分析收集到的数据,将被试分成三个潜在类别,潜在类别概率分别是0.5487,0.3914,0.0599,根据其数据分布和结构特点,分别将它们命名为:"通勤花费敏感型"、"看重工作型"、"注重体验型",并得出个类别下的条件概率估计值。最后对每一个类别下的选择异质性进行分析。
[Abstract]:Commuting is not only the most important trip for urban residents, but also the most important travel. It has a decisive impact on the daily traffic conditions of the city. The most significant performance of urban traffic congestion is the traffic congestion during commuting hours, which is limited by urban traffic resources. Expanding the supply of roads and improving the quality of transportation services cannot be taken as a permanent measure. The discrete choice model is a classical way to study the path choice and mode choice of commuters, and its development process is improved with the demand of people. The hybrid Logit model can be used to predict the selection pattern according to the individual's random preferences, while the potential category model, which is focused on the further study of unmeasurable factors, to a large extent, controls variables. The association between explicit variables that need to be examined is measured by potential variables, and a new classification of a population is made at the same time. We can also find out the proportion of this group in the crowd, which is conducive to our further study of the subject. The potential class model method is an important supplement to the study on the heterogeneity of commuter travel behavior. The specific contents and main conclusions of this paper are as follows: first of all, Through a new understanding of commuting trip, we can understand the spatial characteristics, intensity characteristics, purpose characteristics of urban commuting trip. Secondly, the achievements in the field of travel behavior selection based on traditional model are realized, and it is recognized that, in many cases, The methods of adding roads, elevated structures and tunnels have obvious effect only in the short term. Then introduce and analyze the mathematical principle of the model of potential categories, and use several basic examples to distinguish the difference between the existing classification and the subdivision of the population. Finally, Collecting data by declarative survey method, classifying the surveyed population by using the potential category model, obtaining the parameter estimation and influence factor coefficient of each attribute, naming each category, analyzing the characteristics of each category. Weather conditions, commuting distance, commuting costs, attendance matters, comfort, identity were selected to do declarative surveys, each attribute selected was divided into three levels, the data collected through LatentGold software analysis, The subjects were divided into three potential categories. The probability of potential categories was 0.54870.39140.5999. According to their data distribution and structural characteristics, They are named as "commute cost sensitive type", "emphasis on work type", "emphasis on experience type", and the conditional probability estimates under each category are obtained. Finally, the selection heterogeneity under each category is analyzed.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F572

【参考文献】