SMART International Symposium for Next Generation Infrastructure: An integrated residential and transport mobility modelling framework using agent-based technology
A presentation conducted by Mr Matthew Berryman, SMART Infrastructure Facility, University of Wollongong. Presented on Tuesday the 1st of October 2013.
Modelling and analysis of large systems of infrastructure systems carries with it a number of challenges, in particular around the volume of data and the requisite
complexity (and thus computing resources required) of models. In this paper we discuss both some novel architectures for scalability of modelling as well as for fusion and relevant visualisation of large data sets. We have a particular focus on geospatial infrastructure data visualisation.
Semelhante a SMART International Symposium for Next Generation Infrastructure: An integrated residential and transport mobility modelling framework using agent-based technology
Semelhante a SMART International Symposium for Next Generation Infrastructure: An integrated residential and transport mobility modelling framework using agent-based technology (20)
How to Troubleshoot Apps for the Modern Connected Worker
SMART International Symposium for Next Generation Infrastructure: An integrated residential and transport mobility modelling framework using agent-based technology
1. ENDORSING PARTNERS
An integrated residential and
transport mobility modelling
framework using agent-based
technology.
www.isngi.org
The following are confirmed contributors to the business and policy dialogue in Sydney:
•
Rick Sawers (National Australia Bank)
•
Nick Greiner (Chairman (Infrastructure NSW)
Monday, 30th September 2013: Business & policy Dialogue
Tuesday 1 October to Thursday, 3rd October: Academic and Policy
Dialogue
Presented by: Matthew Berryman SMART Infrastructure Facility,
University of Wollongong
www.isngi.org
2. An integrated residential and transport
mobility modelling framework using
agent-based technology.
Dr Matthew Berryman
7. Options for scalability
• Run on an HPC (large numbers of agents, MPI
between nodes), but:
– At the time the project started, no HPC support in
TRANSIMS;
– Limited team skill sets in HPC; and
– No need at that stage as we are looking at a
subregion of Sydney. Instead, seemed best to
achieve speedups by using:
• Multiple scenario/seed runs distributed across
a cloud, with a central database.
– Still some work to integrate, and to automate
deployment of multiple model VMs.
9. TRANSIMS integration
• Our main goal was to extend our agents to
have travel ability through use of a
microsimulator.
• Need to maintain a one-to-one mapping
between agents in our model, and their
TRANSIMS representation.
• Used only the router and (initially) the
microsimulator from TRANSIMS; do the rest
inhouse (in REPAST Simphony [sic]).
10. TRANSIMS inputs
Our software supplies to TRANSIMS an agent’s
• ID,
• the household ID that they belong to,
• the purposes of the trip (go to home, go to work, go to school,
go shopping, go for social recreation or other purposes),
• the travel mode of the trip (for instance car, bus, train, bicycle,
walk, or using carpool as a car passenger), the start time and
expected arrival time of the trip,
• the origin and destination location of the trip.
If the agent travels by car, they will also need to provide for
TRANSIMS:
• which car in the house they are using (for instance the second
car in the house), and
• where they park that car as well.
11. TRANSIMS outputs
• Based on these output data, the Sydney model
collects the travel time of each trip, using them to
calculate the travel cost of the trip by using the
current travel mode and other travel modes.
• Agents, based on these costs, make their own
decision about their travel mode for their trips in
the next time step. Our model also utilises the
congestion statistics from TRANSIMS output to
calculate the satisfaction for agents to make a
decision of relocation (staying or moving out the
study area).
12. Lessons learned
• Having a single, efficient data structure is essential
for having easy to maintain and bug free code. Load
data from the database and then use that at the
central point of view. There is a need to handle birth
and death processes for individuals and family
groups, and have those reflected across central data
structure, output database, and TRANSIMS, which
was a bit painstaking.
• Dropped the microsimulator—running the router
only is sufficient—bearing in mind our need is for
relative travel times across modes and between
travel zones (several blocks large), as well as for fast
running times.