ABSTRACT

1
Rutgers Intelligent Transportation Systems (RITS)
Laboratory Department of Civil Environmental
Engineering
Modeling Traveler Behavior via Day-to-Day
Learning Dynamics Impacts of Habitual Behavior
Paper No 10-2607

Ozlem Yanmaz-Tuzel, M.Sc. And Kaan Ozbay,
Ph.D. Rutgers, the State
University of New Jersey
LITERATURE REVIEW

• METHODOLOGY
• In the proposed framework, an agent-based
  learning system via Bayesian-SLA is designed
  which can learn the best possible actions and
  model travelers day-to-day travel choices in a
  non-stationary stochastic environment.
• Each traveler maintains a choice probability
  profile for the available alternatives, and
  updates his/her probability profile based on
  previous travel choices, exhibiting a tendency to
  search for satisfying choices rather than the
  best behavior inertia and bounded rationality.
• The estimated learning parameters reflect
  travelers perception about the system and their
  response to the experienced traffic conditions.

ABSTRACT This paper focuses on the development
of learning-based behavioral mechanisms for
updating route and departure time choices when a
major and brand new facility is added to the
existing transportation system by creating new
route choices that did not exist previously. To
model this complex user behavior based on
empirical observations this study applies the
Bayesian-SLA framework recently developed by the
authors. In this approach, Bayesian-SLA framework
systematically accounts for commuters belief,
perceptions and habitual tendencies about the
transportation system, and represents these
dynamics as random variables. The developed
learning model is calibrated and validated using
real traffic and travel time data from New Jersey
Turnpike (NJTPK) toll road to investigate the
impacts of Interchange 15X installation on the
day-to-day departure time and route choice
behavior of NJTPK travelers. The estimation
confirm strong effect of habitual behavior on
traveler choice, consistent with the preliminary
traffic volume analysis findings. The proposed
Bayesian-SLA model can successfully capture the
significant learning dynamics, demonstrating the
possibility of developing learning models as a
viable approach to represent travel behavior.

• INTRODUCTION
• Modeling and understanding the relationship
  between individuals travel perception and
  learning process and the day-to-day traffic flows
  remain an important challenge for transportation
  researchers.
• This paper focuses on the development of
  learning-based behavioral mechanisms for updating
  learned route and departure time choices in the
  presence of new route inclusions to the
  transportation system considering the impacts of
  habitual behavior on travelers choice
  mechanisms.
• This study applies the Bayesian-SLA framework.
  In this approach, Bayesian-SLA framework
  systematically accounts for commuters belief,
  perceptions and habitual tendencies about the
  transportation system, and represents these
  dynamics as random variables.
• The developed learning model is calibrated and
  validated using real traffic and travel time data
  from New Jersey Turnpike (NJTPK) toll road to
  investigate the impacts of Interchange 15X
  construction on the day-to-day departure time and
  route choice behavior of NJTPK travelers.

• EMPIRICAL SETTING
• The developed Bayesian-SLA framework is
  implemented to investigate the impacts of the
  addition of 15X Interchange on December 2005 on
  the day-to-day departure time and destination
  choice behavior of NJTPK travelers.
• After nearly three years of construction, NJ
  Turnpike Authority (NJTA) opened the 250 million
  Interchange 15X on the Eastern Spur (just south
  of Interchange 16E) on December 1, 2005. The new
  interchange serves the new Secaucus Junction rail
  transfer station.
• The NJTA contributed an additional 84 million
  to develop the 450 million adjacent Allied
  Junction, which will have 3.5 million square feet
  of combined commercial and residential
  development, as well as up to 2,600 new parking
  spaces when the development is completed. Upon
  full development, Interchange 15X is expected to
  handle 40,000 vehicles per day.

2
Rutgers Intelligent Transportation Systems (RITS)
Laboratory Department of Civil Environmental
Engineering
Modeling Traveler Behavior via Day-to-Day
Learning Dynamics Impacts of Habitual Behavior
Paper No 10-2607

Ozlem Yanmaz-Tuzel, M.Sc. And Kaan Ozbay,
Ph.D. Rutgers, the State
University of New Jersey

• Estimation Results
• The estimation process resulted in Beta
  distribution for the posterior distribution of
  each parameter. Since beta distribution always
  lies within 0, 1, the constraints on the
  learning parameters will be satisfied at all
  times.
• Mean values for the parameters (a, b) are (0.029,
  0.0029), and standard deviations are (0.011,
  0.00093), respectively.

• The prior distribution of the learning parameters
  (a, b) can be represented by p(a,b). In this
  paper, Dirichlet (multivariate generalization of
  the beta distribution), multivariate Normal and
  Uniform distributions are tested as joint prior
  distributions of p(a,b)
• The posterior distribution of the learning
  parameters given the observations, p(a,bD), can
  be calculated using Bayes theorem
• Posterior distribution of the learning
  parameters is a very complex multidimensional
  function which requires integrating. Thus, to
  obtain the mean and variance of the parameters
  (a, b) Metropolis-Hastings (M-H) algorithm is
  used. The M-H algorithm is a rejection sampling
  algorithm used to generate a sequence of samples
  from a probability distribution that is difficult
  to sample from directly.
• To ensure MCMC convergence, Heidelberger and
  Welch first test diagnostic was employed. This
  diagnostic compares the observed sequence of MCMC
  samples to a hypothetical stationary
  distribution, using the Cramer-von-Mises
  statistic. The test iteratively discards the
  first 10 of the chain until the null hypothesis
  is not rejected (i.e. the chain is stationary),
  or until 50 of the original chain remains

• CONCLUSIONS
• This research focuses on modeling learning based
  behavioral mechanisms for updating route and
  departure time choices in light of new facility
  additions to the existing transportation system.
  The proposed model extends the existing SLA
  theory by using it in a Bayesian framework and
  bounded rationality (BR), while considering the
  impacts of habitual behavior
• Day-to-day learning behavior is modeled based on
  Bayesian-SLA theory, where each user updates
  his/her choice based on the rewards/punishments
  received due to selected actions in previous
  days. A linear reward-penalty reinforcement
  scheme is considered to represent day-to-day
  behavior of NJTPK users as a response to
  construction of a new Interchange.
• In order to account for travelers resistance to
  switch routes, concept of habitual behavior
  (inertia) is included in the proposed model, such
  that the travelers switch to the new route only
  if it has significantly less cost.
• Finally, learning parameters were modeled as
  probability distributions rather than
  deterministic values, and Bayesian posterior
  distributions are estimated.
• The empirical results obtained from the real
  transportation network confirm the strong effect
  of habitual behavior on traveler choice.
• The proposed Bayesian-SLA model can successfully
  capture the significant learning dynamics,
  demonstrating the possibility of developing a
  psychological framework (i.e., learning models)
  as a viable approach to represent dynamic travel
  behavior.
• A possible extension of the proposed methodology
  is to investigate how individuals tolerance
  level, and learning parameters change over time
  as the users gain more experience with the
  transportation system.

• Calibration and Validation of the Learning
  Parameters
• Calibration and validation are important
  processes in the development and application of
  day-to-day DTA models. These processes are
  developed to ensure that the models accurately
  replicate the observed traffic condition and
  driver behavior.
• Model calibration is a process whereby the values
  of model parameters are adjusted so as to match
  the simulated model outputs with observations
  from the study site. It is usually formulated as
  an optimization problem to determine the best set
  of model parameter values in order to minimize
  the discrepancies between the observed and
  simulated values.
• The calibration process is then to modify the
  values of the model parameters ?, so to find the
  best set of values which minimizes F. The
  proposed objective function F minimizes the
  difference between observed and simulated
  volumes
• After determining the optimal set of parameters
  from the calibration process, a validation
  process is performed in order to determine
  whether the simulation model replicates the real
  system. Mean standard errors (MSE) are
  calculated for each day the validation process