1 / 20

Micro-simulation of learning and adaptation in activity-travel choice

Micro-simulation of learning and adaptation in activity-travel choice. Background and Objectives (1). Activity based models of travel demand aim to predict: which activities are conducted when where for how long transport mode Examples of existing activity-based models:

Download Presentation

Micro-simulation of learning and adaptation in activity-travel choice

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Micro-simulation of learning and adaptationin activity-travel choice

  2. Background and Objectives (1) • Activity based models of travel demand aim to predict: • which activities are conducted • when • where • for how long • transport mode • Examples of existing activity-based models: • Albatross (operational on a national scale) • Aurora (under development) • Bhat 1999 • Bowman and Ben-Akiva • PCATS

  3. Background and Objectives (2) • Existing models implicitly assume that activity-travel patterns are static • Adaptation refers to a choice to change an existing pattern in response to changed conditions in the environment, household, life cycle of the individual • Learning follows from the fact that individuals have limited knowledge of their environment (locations and routes) • Adaptation and learning means that activity patterns are in a constant state of change

  4. Background and Objectives (3) • The objective is to develop a micro-simulation system of dynamics of activity-travel choice

  5. Approach • Micro-simulation: • Individuals are individually represented as agents • The scheduling and implementation of activities are simulated in space and time • High resolution of space and time • Components of agents: • An activity-based model: Aurora • Spatial cognition: Belief networks and Bayesian learning • Social interaction: Social networks and communication

  6. Bayes theorem Where: Pr(xj | Y) the belief that X = xj given finding Y Pr(Y | xj) the probability of Y given X = xj. Pr(xj) the a-priori belief that X = xj

  7. Bayes theorem: Example • Pr(yes) = 0.3 • Pr(Y | xj) • Pr(yes | Y = no) = 0.2 x 0.3 / (0.2 x 0.3 + 0.9 x 0.7) = 0.087 • Pr(yes | Y = yes) =0.8 x 0.3 / (0.8 x 0.3 + 0.1 x 0.7) = 0.774

  8. Advantages of the Bayesian learning approach • Learning is incremental so that at any moment in time beliefs are defined as a set of probability distributions • The probabilities have a ready interpretation as beliefs that certain locations/routes have certain attribute values • Utilities can be redefined as expected utilities as the criterion for choice • The probabilities can be used to define an entropy measure of the expected information gain that can be obtained from certain activities and trips

  9. Information gain measure (1) • The information quantity of the finding that a cell i has a value xij on some variable Xi is: Bits • The expected amount of information required for identifying the value of a cell is defined as a weighted sum of this measure across values of Xi:

  10. Information gain measure (2) • The info(i) after some observation Yiis: • The expected info(i) beforesome observation is: • The expected information gainis: info(i,) – info(i)

  11. Spatial Search • Information gathering can be modeled as the choice of a route that maximizes the information gain within a given acceptable distance

  12. Individual’s observations (1) • Sensitiveness of an observation is a function of: • The distance from the cell • The nature of the variable observed • Transport mode • Link type • The purpose of the trip

  13. Individual’s observations (2) • Zero sensitiveness is defined as (n is nof levels of y): • Pr(yk | xj) = 1/nk, j • And maximum sensitiveness as: • Pr(yj | xj) = 1 en Pr(yk | xj) = 0, kjj

  14. Individual’s observations (3) • Sensitiveness in general as: • Where • jk are observation-bias parameters determining the probability of Y = yk given X = xj •  is an observation-sensitiveness parameter

  15. A spatial belief network: Variables (1) • Structural type • Inner city • High density urban • Low density urban • Rural • Relationship with road network • Nearby link of main road • Nearby link of local road • No relationship

  16. A spatial belief network: Variables (2) • Main Landuse • Industry • Housing • Shops • Etc. • Availability of facilities for activity A • Attractiveness for activity A • High • Medium • Low • Zero

  17. Land use Neighbor of i Type i Network i Observation N Land use i Observation L Facilities i for A Observation F Attraction i for A Observation A A spatial belief network: Structure

  18. Social Networks • The probability that A is a social contact of B is a function of: • Overlap in action space between A and B • Overlap in socio-economic space between A and B • Unobserved factors • Definition: A is a social contact of B iff • Beliefs/preferences of B are conditionally dependent on beliefs/preferences of A

  19. Communication • A social contact can be modeled as a link between belief networks of A and B • A conditional probability table defines the social contact: • Partial adjustment, biased adjustment of beliefs/preferences • Attribute specific impact • A symmetrical impacts • Information gain of the social contact

  20. Conclusion • Project is still in an early stage of conceptualization • Simulation studies will be conducted to test the validity of the ideas • Data collections are needed to estimate parameters of belief- network model and social-network model • The Aurora, Belief-network and Social-network model need to be implemented as methods of agents

More Related