Transportation and Urban Design Studio E 2020

for Graduate School of Civil Engineering in 2020 Autumn Semester 1 (A1)

Lecturers

• Prof. Eiji Hato (Professor, Transport Research and Infrastructure Planning (TRIP) Lab.)
• Prof. Takamasa Iryo (Professor, Transport Research and Infrastructure Planning (TRIP) Lab.)
• Assoc. Prof. Ryuichi Shibasaki (Associate Professor, Department of Systems Innovation, School of Engineering)
• Assist. Prof. Muhammad Awais Shafique (Assistant Professor, University of Central Punjab)
• Assist. Prof. Junji Urata (Assistant Professor, Transport Research and Infrastructure Planning (TRIP) Lab.)

Place and Time

Online, on Monday and Thursdays 13:00-14:45. Please check the online URL from ITC-LMS

Purpose and Contents of the Course

This course focuses on learning some of methodologies to analyze transportations and regions, which are sometimes vulnerable to natural hazards. In addition to it, getting used to the essence of the basic way of theoretical and mathematical thinking in planning is another main target. For fulfilling these purposes, we choose four topics: 1) Traffic flow modelling, 2) Logistics management and analysis, 3) Statistics and Machine Learning, 4) Travel behavior modelling.
Syllabus

Schedule of the Course

• Sep. 28 Introduction [PDF]

Topic 1: Transportation Modelling and Statistics, by Prof. Urata and Prof. Hato

• Oct. 01 Travel Behavior Modelling (1)
• [Abstract] This talk explains a basic framework of the discrete choice. The framework is composed of four factors: decision-maker, alternative, attribute, and decision rule. When decision-makers choose based on their utility based rule, how to define the utility is essential. Usually, the utility function defines as a linear combination of alternatives' attributions and decision-makers' characteristics. The choice is stochastic if the utility includes an error component because of an analyst's imperfect knowledge. The random utility model gives the formulations of the discrete choice and introduces an (observed) systematic utility and an (unobserved) random utility. The Logit model applies Gumbel distribution as the random utility. This talk derives the equations of the probability of the multinomial logit (MNL) model using the properties of Gumbel and Logistic distributions. The MNL model shows a property of Independence from Irrelevant Alternatives (IIA). We can apply the nested logit model to avoid the IIA problem when the error components of alternatives correlate.
• Oct. 05 Travel Behavior Modelling (2)
• [Abstract] The nested logit (NL) model can relax the IIA property of the logit model. The relaxation is efficient when the alternatives are correlated. The NL model's formulation is composed of the conditional probability when a nest is selected and the marginal probability of choosing an alternative in a nest. The parameters of the utility function are estimated using the Maximum Likelihood Estimation (MLE) method. The approach gets more fit parameters for choice results. We evaluate the goodness of fit of the estimated models by likelihood ratio index and t-statistics. Elasticity is useful to know how the probabilities change in response to changes in associated attributes' value.
• Oct. 19 Presentation from students
• [Abstract] The students had presentations about that they analyzed the transportation behavior and mode choice in Yokohama city using Yokohama Probe Person data 2009. Group A run the nested logit model, which provided a negative parameter of the trip fare. The train fare changes are more effective in people's behavior than bus fare by their elasticity analysis. Group B focused on usable/non-usable times during a displacement. They introduced three nest-structures: walk&bike, bus&rail, car, and got a suitable estimation result. Group C introduced tiredness on the utility functions by using travel distances of going on one's own. Their policy simulation considered a new BRT route, and they evaluated one-lane occupancy by BRT. Group D focused on personal attributes and clarified that gender and age influence to choose a car/bike.

Topic 2: Traffic Flow Theory and Modelling, by Prof. Iryo and Prof. Hato

• Oct. 08 Traffic Flow Theory and Modelling (1)
• [Abstract] The lecture explained about traffic flow theory and mainly focus on dynamic traffic flow and road congestion on a single road. The lecture introduced Stationary traffic flow, Fundamental diagram, Kinematic wave theory, and Shockwave theory. We understood why congestion occurs and how the congestion grows.
• Oct. 12 Traffic Flow Theory and Modelling (2)
• [Abstract] The first half of the lecture talked about the Cell transmission model (CTM) to explain the dynamics of traffic flow. The exercise of the CTM model enables us to calculate free-flow, queue spillback, and bottleneck. The second half of the lecture explained about network traffic assignment problems. The methodology assigns origin-destination traffic volume to each link of the road network, and we can know how much each link is used and congested.
• Oct. 15 Traffic Flow Theory and Modelling (3) and Final Exercise
• [Abstract] The network assignment problem has several methodologies to assign OD traffic volume: User-Equilibrium assignment (UE) and System Optimal assignment. The talk explained how to calculate the UE assignment. The lecture showed an optimization problem equivalent to the UE condition, and the FranK-Wolfe algorithm obtains a solution by an iterative procedure.

Topic 3: Statistics and Machine Learning, by Prof. Shafique and Prof. Hato

• Oct. 22 Introduction to Machine Learning
[Abstract] We will study what Machine Learning is. Where is it used and how it is affecting transportation engineering sector. Some of the popular algorithms will be discussed in detail such as Neural Networks, Support Vector Machines, Decision Trees, Random Forest etc. R packages linked to such algorithms will also be discussed.
• Oct. 26 Machine Learning vs. Discrete Choice Modelling
[Abstract] In this lecture, we will review some of the studies which have done work on the comparison of Machine Learning with conventional Discrete Choice Models to predict travel behavior. Specifically, we will discuss the use of logit models for mode choice modelling. Class exercise will follow.
• Oct. 29 Travel Mode Detection using Machine Learning
• [Abstract] We will investigate the use of sensors' data for prediction of travel mode used by the device carrier. Data cleaning, pre-processing and pro-processing methods will also be discussed. Class exercise will follow.

Topic 4: Logistics Management and Analysis, by Prof. Shibasaki and Prof. Hato

• Nov. 02 Global Logistics Modelling & Analysis (1)
• Nov. 05 Global Logistics Modelling & Analysis (2)
• Nov. 09 Presentation from students

Schedule of the Course

Assignments in each of the three topics (22 points @ 4), and attendance points for classes (12 points)
Your attendance is confirmed using the history function of zoom.