- : Ms Word, Ms Word Format
- : 100 Pages
- : ₦5000
- : 1-5 Chapters
- Click to DOWNLOAD Materials
A MICROSIMULATION MODEL OF ACTIVITY PATTERNS AND WITHIN HOUSEHOLD INTERACTIONS
ABSTRACT
Activity-based approaches to travel demand analysis have gained increasing attention over the past decades. This thesis aims to propose an algorithm that belongs to this field. Its objective is to simulate individual’s daily activity-travel patterns, and to incorporate the interactions among members of particular household. This model uses several tools to simulate the activity patterns. A lot of effort has been dedicated to cluster analysis of activity patterns. The methodology has been studied and an approach that suits this problem is suggested. Decision trees, particularly the CHAID algorithm, are used to take into account the personal and household characteristics.
The individual’s daily activity patterns that carry information about activity participation on a detailed temporal scale are the output of this model. These patterns respect individuals’ constraints that are implicitly embedded. The model was evaluated using data from the Centre County, Pennsylvania that were collected during fall 2002 and spring 2003.
TABLE OF CONTENTS
Chapter 1 OVERVIEW…………………………………………………………………………………..1
Thesis structure………………………………………………………………………………………..4
Chapter 2 TRANSPORTATION PLANNING AND THE PROPOSED
ALGORITHM …………………………………………………………………………………………5
2.1 Urban Transportation Planning System (UTPS) ……………………………………..5
2.1.1 Trip Generation …………………………………………………………………………6
2.1.2 Trip Distribution………………………………………………………………………..7
2.1.3 Modal Choice ……………………………………………………………………………8
2.1.4 Traffic Assignment…………………………………………………………………….8
2.1.5 Some Drawbacks of the UTPS…………………………………………………….9
2.2 Activity-Based Approach to Travel Demand Analysis…………………………….11
2.2.1 Utility Maximizing Models…………………………………………………………16
2.2.2 Computational Process Models (CPM)…………………………………………19
2.3 The algorithm developed in this thesis…………………………………………………..20
2.3.1 Framework of the proposed model……………………………………………….22
2.4 Summary……………………………………………………………………………………………29
Chapter 3 THE DATA SET DESCRIPTION AND DATA REPRESENTATION …31
3.1 Data collection……………………………………………………………………………………31
3.2 Statistical description of the data set ……………………………………………………..35
3.3 Generating of synthetic schedules …………………………………………………………38
3.4 Representation of within household interactions……………………………………..41
3.4.1 The entire household and not an individual must be included in the
analysis………………………………………………………………………………………41
3.4.2 Variable reflecting interactions will be included…………………………….42
3.4.3 The data files for clustering…………………………………………………………45
3.4.4 Representation of within household interactions ……………………………46
3.5 Summary……………………………………………………………………………………………52
Chapter 4 METHODOLOGY FOR DATA ANALYSIS AND SIMULATION……..54
4.1 Identifying groups in data…………………………………………………………………….54
4.1.1 Hierarchical clustering………………………………………………………………..57 4.1.2 Partitioning clustering ………………………………………………………………..59 4.1.3 K-medoids algorithm………………………………………………………………….61 4.1.4 How to determine the right number of clusters? …………………………….63
4.1.5 Dissimilarity measures ……………………………………………………………….65
4.2 Genetic algorithms for clustering ………………………………………………………….68
4.2.1 Brief introduction to genetic algorithms………………………………………..68
4.2.2 Adapting GAs for k-medoid clustering…………………………………………72
4.2.3 Why should we use GA?…………………………………………………………….77
4.3 Link between socio-demographic characteristics of individual and
households to groups of activity patterns………………………………………………78
4.3.1 Modeling tools…………………………………………………………………………..78
4.3.2 Comparison of multinomial logit model to decision trees ……………….82
4.3.3 Action assignment in decision trees ……………………………………………..84
4.4 Simulation of the daily activity patterns (activity assignment model)………..87
4.5 Evaluation of results……………………………………………………………………………89
4.5.1 Comparison of activity profiles……………………………………………………89
4.5.2 Comparison of time spent in activities………………………………………….92
4.5.3 Comparison of number of episodes………………………………………………93
4.6 Summary……………………………………………………………………………………………94
Chapter 5 APPLICATION AND RESULTS……………………………………………………..96
5.1 Cluster Analysis………………………………………………………………………………….96 Parameter settings:……………………………………………………………………………..97 The output:………………………………………………………………………………………..98
5.1.1 Results for one-adult households………………………………………………….101
5.1.2 Results for two-adult households in which the head of the family
works full time ……………………………………………………………………………102
5.1.3 Results for two-adult households in which the head of the family does not work full time (other than full time)………………………………….104
5.1.4 Results for three-adult households ……………………………………………….105
5.1.5 Summary of cluster analysis………………………………………………………..106
5.2 Decision Trees……………………………………………………………………………………107
5.2.1 Implementation and parameter setting of decision trees………………….107
5.2.2 One-adult households…………………………………………………………………109
5.2.3 Two-adult households, head of the households employed full time….113
5.2.4 Two-adult households, head of the households not employed full
time……………………………………………………………………………………………114
5.2.5 Three-adult households ………………………………………………………………114
5.2.6 Summary of the CHAID analysis…………………………………………………114
5.3 Probabilistic action assignment …………………………………………………………….115
5.4 Simulation and Evaluation……………………………………………………………………117
5.4.1 Determination of the best representation of within household
interactions …………………………………………………………………………………118
5.4.2 Overall performance of the simulation model………………………………..122
5.5 Summary of the fifth chapter………………………………………………………………..135
Chapter 6 CONCLUSIONS AND FUTURE RESEARCH………………………………….136
6.1 Overview……………………………………………………………………………………………136
6.2 Accomplishments ……………………………………………………………………………….136
6.3 Summary……………………………………………………………………………………………138
6.4 Future research……………………………………………………………………………………139
Bibliography ………………………………………………………………………………………………….141
Chapter 1
OVERVIEW
The traditional goal of modeling in the field of travel demand analysis is to estimate the volume of traffic on particular roads in a transportation network (Ortuzar and Willumsen 1994). Recent policy propositions and actions (e.g., introduction of new technologies, taxation, and pricing of congestion) and market trends (e.g., market penetration of mobile telecommunication technologies) also motivate the study of impacts not only on traffic volumes but also car ownership, trip consolidation into chains, departure times, and more general shifts in spatial and temporal aspects of travel demand. For this reason, much effort has been dedicated to create models that would address travel behavior in a more comprehensive way looking at schedules of activities to assess the impact of policy actions. The required precision and level of detail of such models has also changed dramatically over the years. An estimate of only daily volumes was sufficient in the early stages of modeling and regional simulation. However, the current objective of research is not only to find a model that can estimate traffic volumes at much smaller time scales but to also estimate other changes in travel behavior, such as departure time shifts affecting peak spreading and changes in time allocation from weekdays to weekends, that may create unseen congestion types. It is clear that this is not an easy task and that the original model (UTPS) developed in early 1950s is not sufficient (a discussion of its limitations is provided in Chapter 2).
A way to overcome these limitations is by using activity-based approaches to travel demand analysis. The idea behind these approaches is that travel demand is derived from demand for activities. The models attempt to estimate the sequence of activities an individual follows in a day, called a synthetic schedule or activity pattern. Once the schedule is known, the derivation of the volumes on particular roads in the network is a rather straightforward algebraic operation.
The remaining problem to solve is the creation of a model that estimates the activity patterns of individuals. However, the way in which an individual creates his schedule is very complex. It is not feasible to list all the factors that affect the individual’s planning of the activity schedule; however, the following list contains at least the more important of them:
An individual does not plan the whole schedule at one time instant. Scheduling is a continuous process with multiple time horizons.
There are many constraints that affect the decision making process, such as interactions with the environment or other individuals in the study area. The schedules have to reflect individuals’ needs for participating in activities, the performance of the transportation system, availability of transportation supply, and many other issues.
It is a very difficult task to measure these constraints. These constraints may vary between individuals and might be purely subjective.
Two individuals facing the same constraints may, and probably would, develop different schedules.
The process is not exactly an optimization problem (Simon 1983). People usually do not compare all possible schedules and pick the best one. They often tend to use one that has worked in the past, even though there could be another schedule that is “more optimal” from an objective point of view.
There are many factors in the schedule that need to be specified, for example the number of activities to be scheduled and their sequencing, the timing, duration, and company of each activity, travel mode to and from this activity, and others. The dimensionality of such a problem is very high (see “The fundamental modeling problem” later in this thesis for an example and explanation).
There are many different aspects of the schedules that must be determined. Most of the models proposed need to assume a hierarchy of causality (nested logit is often used, for example in Wen and Koppelman (2000)). For example, does a person first decide on the timing of an activity, which further conditions its duration, or the other way around (Pendyala 2003)? What about activity sequencing, travel mode used and other factors? Previous research has not succeeded in clearly determining this causality (Doherty 2003) even though it is essential for the performance of many models.
This list of a few issues should demonstrate the complexity is faced. The modeler is dealing with human behavior and decision-making processes. The examples listed above should also give the impression that simplification is necessary for any modeling attempt.
Much effort has been spent on this problem, and knowledge in the field has developed significantly in just a few years (for an overview see Arentze and Timmermans 2000). Many different models that have been proposed are described in this dissertation. There is no single model that considers all of the issues that a real human being considers during her/his activity planning procedure.
The research in this dissertation introduces a microsimulation model that generates activity patterns of individuals in a given study area. The proposed algorithm belongs to the field of activity-based approaches to travel demand analysis. It aims to replicate the observed patterns, which implicitly include the constraints and the outcome of the decision making processes underlying a person’s time allocation in a day with other persons and alone. By replicating the entirety of a person’s activity-travel pattern in a day, a feasible and robust solution that consists of the timing and sequencing of activities is provided. These synthetically generated schedules are also linked to individuals and household’s characteristics (such as income, number of cars, number of children, age, gender, and others).
The main contribution in travel behavior research of the proposed approach is in capturing the interactions among household members. For example, in a family with a child of school age, one of the parents has to adjust her/his schedule to be able to drop off the child in the morning at school. Another example can be a joint dinner of both parents.
Both must adjust their schedules to meet at the same time and at the same restaurant. These examples show the importance of such a model. The second important contribution is a new method for clustering that uses genetic algorithms paving the way to a variety of applications in travel behavior. A variety of other algorithm propositions are also offered and demonstrated that can be used alone or in tandem with other models. The objectives defined at the early stages of research were met as expected. However, many avenues were left as future tasks and are reviewed in chapter 6.
Thesis structure
The thesis has the following structure. The purpose of the second chapter is to introduce the field of transportation planning and define terms that are essential for understanding the algorithm proposed in this thesis. Also, a review of literature that is connected to the thesis is provided. This chapter also contains a basic description of the proposed algorithm. The third chapter focuses on the data that are used in this thesis. First the data collection method is briefly described and then the representation used in this thesis is explained. The fourth chapter provides detailed data analysis and simulation method descriptions. First an introduction to cluster analysis is provided, then the decision tree methodology is presented. At the end of this chapter, the approach to simulation used in this thesis as well as methods of evaluation of results are discussed. The fifth chapter describes the application of particular phases of the algorithm using a step-by-step approach. For every topic, the parameter settings as well as the results are discussed. The end of this chapter provides the discussion of findings and evaluation of the model. Finally, the sixth chapter concludes the thesis, provides a summary and discussion of results, and last but not least, suggests some future steps that would improve the performance of the algorithm.
