UNIVERSITY
OF CALIFORNIA TRANSPORTATION CENTER
FINAL
REPORT
PROJECT
TITLE: Activity-Based Forecasting Model for Planning
Applications
PRINCIPAL
INVESTIGATOR: Will Recker
DEPT
& CAMPUS ADDRESS: Institute of Transportation Studies
University
of California, Irvine
Irvine,
CA 92697-3600
TELEPHONE: (949)
824-5642
FAX
NUMBER: (949) 824-8385
E-mail Address: wwrecker@uci.edu
FUNDING: UCTC Year 13 Research Grant
OVERVIEW:
The work completed here is based on previous
activity-based research conducted by the principal investigator and his
colleagues and is directed toward developing a practical planning application
of a mathematical programming activity-based model as an effective travel
demand forecasting tool.
In this research, we seek to complete the
modeling framework that has evolved over past research efforts by extending it
to a “traditional” planning framework.
Specifically, we couch the activity-based approach in terms that are
amenable to its development as a planning tool for travel demand forecasting
that not only provides output consistent with accepted trip-based static
planning methodologies, but further provides full estimates of the associated
dynamics of trip generation, distribution and route selection; all from a
theoretically consistent paradigm based on the need/desire of households to
interact with their environment. By
showing that the particular mathematical programming paradigm can be used to
describe the demand modeling processes both for conventional trip-based travel
demand and for activity-based approaches it is hoped not only to facilitate the
practicality of activity-based modeling approaches, but also to tap into the
wealth of research that has guided mainstream travel demand analysis.
KEY FINDINGS:
In this research, we establish the foundation for employing the
Household Activity Pattern Problem (HAPP) model (a mathematical programming
activity-based model offered by Recker, 1995) within a planning context. The HAPP model is in the form of a Mixed
Integer Linear Programming model (MILP), i.e., one comprising continuous
variables (such temporal attributes of an activity pattern as the starting
times of the associated activities) as well as discrete variables (e.g.,
attributes associated with the sequencing of activities, travel modes used, and
persons performing the activities).
The HAPP model system
has evolved, through a series of well-defined stages, to the point that it can
serve as an “estimable” travel demand model that is unified in its approach; it
is specified with inherent coupling across all members of the household, as
well as with inherent coupling of the conventional aspects of travel choice,
i.e., trip generation, travel mode selection, destination, and route. Additionally, its outcome presents
consistent definition of the temporal aspects of all of these aspects for each
household during the analysis period.
As such, conventional trip-based planning models (e.g., the “four-step”
process) theoretically are equivalent to the projection onto the spatial plain
of the results obtained under the HAPP modeling system. In this regard, extension of the HAPP demand
modeling system to a planning model context can provide not only conventional
forecast information (e.g., static O-D matrices, mode shares, and link traffic
volumes) that intrinsically captures the “feedback” process absent in the
four-step approach, but also dynamic traces of these parameters throughout the
course of the “travel day.” Moreover,
because of its microscopic nature, it could logically serve as a candidate
activity model to “front end” the well-known federal TRANSIMS effort that has
thus far had only limited success in wedding demand information to its
microscopic traffic simulation.
Since the HAPP system relies on the type of
microscopic data common to household activity/travel diary surveys, an
essential feature of the research to extend HAPP to a planning context involves
the generation of synthetic households, and their associated diaries, to
constitute the population. This has
been accomplished by first specifying the covariance structure of the
instrumental variables used in the HAPP model system, based on the sample
statistics. Then a stochastic
simulation model, with distributions conditioned by the covariance structure,
is used to generate the population.
Activity/travel patterns of households within this population, with
their associated travel diaries and socio-economic characteristics and based on
the utility specification arising from the estimation of the model, are modeled
using the HAPP system, and results aggregated to a level consistent with the
comparable trip-based models.
The detailed
activity/travel behavior information required on travel and activity
participation for each member, as well as transportation supply information
(including household vehicle holdings and network travel times were drawn from
the Southwest Washington and Oregon Area 1994 Activity and Travel Behavior
Survey, which contains sufficiently detailed information, including
comprehensive travel/activity diaries (with mode availability) and a regional
transportation network model, for an application of the HAPP model.
The sample drawn for
the survey includes 2,232 households and 5,125 persons with a total of 67,016
activities and 37,965 trips (each split fairly equally between two consecutive
survey days). Available household
information includes: household size, household income, type of dwelling unit,
and the number of available vehicles.
The survey also provides person-level data including age, gender,
employment status, occupation, student status, and driver license status.
The activity locations listed in the geocode
file were matched to the different activities of the household by the unique
identification number of the activity.
The duration of the activity and travel were then computed from the
given data for each activity. The
average activity starting and ending times for each activity type were computed
for the whole sample to provide benchmark information on the temporal
flexibility of the activities. Extensive GIS-based (ARC/Info) files for the
area were constructed and include mappings of land use, census demographic
information, and local employment estimates.
Portland Metro made available 1990 Census Tiger files and tract
demographics, as well as EMM/2 coded transportation networks and models. The street address map of the Portland
network is based on an enhanced version of the Census Bureau’s TIGER
files. Shortest path travel times
between all activity locations of a household were generated for all the
households in the sample using TRANSCAD.
This procedure allows for the exploration of all possible
activity/travel linkages for each household, which is fundamental to the
optimization procedure.
The generation of sample
characteristics that were used to specify the covariance structure of the
instrumental variables used in the study included, for example, the
distributions of activities, by type,
together with the distributions of starting,
ending, and duration times of all activities within the sample, e.g.,:
