Background
The
concept of large-scale integrated models has long been recognized as a logical
objective among urban and transportation planners. However, its complexity and
high cost of research and development has also made a record of frustration
that culminated, in 1973, with Douglas Lee’s “Requiem for large scale
models"(8). Our context today is
quite different than in 1973, not just in terms of computational power but also
in terms of data quality and quantity, essential for calibration (9). It is thus without surprise that we
now see a new wave of research in large-scale integrated models.
The common approach is through loose coupling of different
models, each one specialized on a component. The interface between models
consists of exchanging files or API (Application Programming Interface) calls. Majority
of the recent efforts are concentrated towards integrating activity-based
models with dynamic traffic assignment models in an agent-based framework. The
demand component of these integrated models attempt to simulate the daily
activity schedules of each individual (considering travel decision such as day
pattern, mode, route and departure time choice for activities in a day). These
activity schedule then passed on to the supply model, which execute their daily
schedules in relation with the transport network. During the process network
travel times are updated and feedback to the demand component. The iterative
process is used to brings consistency between the two models. Among these
efforts most recent is integration of CEMDAP (an ABM) with MATSIM (10). CEMDAP (Comprehensive Econometric
Micro-simulator for Daily Activity-travel Patterns)(11), focused primarily on activity scheduling. MATSIM (6) on the other hand, in its own assumes
individuals initial plans of the day which are derived on the basis of the
household survey data. These initial plans are then executed in MATsim
demand-supply simulator and, based on the score, agents adapt their plans in
response to conditions that arose during the simulation. The scores within the
MATSim are based on heuristic utility functions with a limited set of
variables, mostly network performance related. The new plans are generated
based on iterative feedback mechanism by modifying few scheduling dimensions of
initial plans in order to get a stable solution, which they called a schedule
user equilibrium (12). The
integration of CEMDAP with MATSIM provide a framework which is more richer in
terms of incorporating behavioural notions of individual travel decisions. The
challenge is then to make these models speak with each other and guaranteeing
full consistency. Moreover, it is difficult to implement just-in-time feedback
processes such as activity rescheduling due to within-day dynamics. For
example, in a major disruption scenario, many agents will need to
reschedule/cancel their upcoming activities on a non-user equilibrium basis,
i.e. with partial awareness of the options and of other agent’s decisions. Another
example of such an integaration is based on the framework named as FEATHERS(13). In a study in the Flanders region
of Belgium, it was connected with ALBATROSS(14),
which provides it with daily activity schedules for the entire population using
a rule based paradigm. Further, it have other components that model
rescheduling decisions, in relation with supply side. Again, some effort is
needed to combine the models together and make them spatially and temporally
consistent(15).
While, in general, the approach has been to combine two or
more sophisticated models on a loosely coupled fashion (e.g. activity-based
demand simulator with a dynamic traffic assignment model), some models exist
that are developed on the notion of integrated framework that can simulate full
day activity/travel schedules. TRANSIMS (16)
is an earlier example of such integrated model. There exist four distinctive
modules in TRANSIMS such as Population Synthesizer, Activity Generator, Route
Planner and Microsimulator. The Activity Generator module in TRANSIMS uses
collected household survey data to work out almost all scheduling dimensions of
activity patterns of synthetic individuals using some rules, random selection
and matching of few socio-economic characteristics of individuals from the
survey. Further within TRANSIMS, there is a feedback mechanism introduced
between Router and Microsimulator, which attempts to bring the system into equilibrium.
However, during that process, individuals can only change their routes with no
flexibility of changing other dimensions of their activity patterns. Later,
MATSIM was developed on similar notion as TRANSIMS, and both have been used
with ABM to overcome the lack of behavioural richness at demand side.
In addition to the efforts of integrating demand and
supply models of day level travel decisions, some efforts are found in the
literature where long- term decision models are loosely coupled with day level
travel decision models. For example, Urbansim combines land-use, demographic
and business establishment models, where agents make long-term decisions (e.g.
home and job relocation) by considering, among others, transport accessibility
(17). A common approach to obtain
accessibility measures is by calling travel models, such as Transcad (18), which runs a 4-step model,
TRANSIMS(19), or MATSim(20), the latter two following an
activity-based paradigm. Very recently, an integrated modeling system, SimTRAVEL
(Simulator of Transport, Routes, Activities, Vehicles, Emissions, and Land) is
presented with a view to more tightly tie together long term and mid-term level
decisions (1). SimTRAVEL claimed to integrate
three separate model in a unified framework, these are Urbansim (landuse
model), OpenAMOS (ABM) (21) and MALTA
(DTA) (22). The design structure of
SimTRAVEL gives much emphasis on integration of OpenAMOS and MALTA, and
Urbansim model is loosely coupled by providing network travel times based
accessibility measure to simulate location choices for the next year. However,
later OpenAMOS is extended to have location choice and vehicle ownership
models. Most of the case studies reported using SimTRAVEL are also based on
application of OpenAMOS and MALTA. Berryman et al (23) argued that absence of microscopic traffic simulator in
SimTRAVEL does not allow to test cases which require capturing of dynamics at a
sufficiently low level of scale, such as impacts of different light rail transit
alignments on traffic and landuse patterns.
Another
notable and very recent effort in the area of integrated framework is POLARIS (Planning and
Operation Language for Agent-based Regional Integrated Simulation), which build
on the notion to provide platform within which several separate models are
plug-in together to provide integrated agent-based platform. So far, the use of
POLARIS is limited to integrate only ABM and DTA models i.e. an integration of
ADAPTS (24) and a mesoscopic traffic
simulation model with a special focus on carrying out intelligent transport
based case studies. Auld et al (2) reported
priliminay results of one such study for Chicago region. According to Hope et
al (25), the design of POLARIS is
such that it incorporates innovative computaitional methodologies to make sure
integration of different models is intact and at the same time provide high
performances. However, the platform is still in development phases and no
notable examples of integrated studies exist that consider long-term level
decisions.
The literature review clearly indicates that integrated
modeling framework is still an emerging field, and so far successful
integration modeling examples are limited to combine ABM and DTA models. The
recent efforts are attempting to integrate long term models with ABM and DTA,
however, the integration is mere on the basis of exchanging data files. This
paper reports development of fully integrated SimMobility framework, that is
developed primarily to integrate Long-term (land use models), mid-term (ABM+DTA)
and short-term (microscopic traffic model) models in unified framework. Within
long-term it utilized Urbansim, at mid-term it utilses day-activity schedule
approach with DYNAMIT as its supply side and the short term is using MITSIM at
its base. The main objective was to provide a platform which enables the
simulation of a portfolio of technology, policy and investment options under
alternative future scenarios, where impacts of the scenario can be visualized
at its three levels in a consistent manner.