SWARMFEST 20071 An Agent-Based Simulation For Emergency Response Management Timothy Schoenharl, R. Ryan McCune, Greg Madey of the University of Notre Dame.

SWARMFEST 2007 1

An Agent-Based Simulation For Emergency Response Management

Timothy Schoenharl, R. Ryan McCune, Greg Madey

of the University of Notre Dame du Lac

This material is based upon work supported by the National Science Foundation under Grant No. CNS-0540348

SWARMFEST 2007 2

Wireless Phone-based Emergency Response System

SWARMFEST 2007 3

Purpose of Simulation Limited knowledge

of events by emergency managers

Understand crisis development

Develop strategy for better response

http://www.nyc.gov/html/fdny/gif/photo_unit/fires/122602/122602_157_truck.jpg

SWARMFEST 2007 4

Overview of Simulation Agent-based simulations model

behavior of city’s population Simulate normal and crisis behavior Output call activity and agent location

on GIS map

SWARMFEST 2007 5

Implementation RePast 3.1 for Java

Colt High Performance Scientific Library

Geotools OpenMap

SWARMFEST 2007 6

Implementation Pattern Oriented Design Singleton Class

SWARMFEST 2007 7

The Agent Each agent represents pedestrian Randomly assigned initial position

inside of Voronoi Cell Call activity based on empirical

distributions Movement determined by given

scenario

SWARMFEST 2007 8

The Environment

Real city created with GIS files Cell phone tower

locations Roads, Water, Political Boundaries

Voronoi cells built around cell phone towers

SWARMFEST 2007 9

Simulation Scenarios Simulate Regular behavior

Move and return Location and call activity based on empirical data

Crisis behavior Flee – Run from a point Bounded Flee – Run a distance from a point

SWARMFEST 2007 10

Validation Movement Model

Face Value Call Activity

Empirical vs. Simulated results plotted

Kolmogorov-Smirnov test

SWARMFEST 2007 11

Validation cont.

SWARMFEST 2007 12

Interface RePast GUI OpenMap Display Voronoi cell color changes with agent

containment Call Activity Graph

SWARMFEST 2007 13

Screenshot of Repast GUI

SWARMFEST 2007 14

Runtime Performance Runtime vs. Graphical Output

500 Agents Move and Return Model Distribution-Based Calling Activity

GIS Agent Location Snapshot Time (s)

No No No 240

Yes No No 354

Yes Yes No 363

Yes Yes Yes 35200

SWARMFEST 2007 15

Runtime Performance Runtime vs. Number of Agents

Linear Scaling demonstrates excellent runtime characteristics

SWARMFEST 2007 16

Summary Agent-based model to simulate crisis

behavior Movement dependent on behavior

model and outputted to GIS map Calling Activity based on empirical data Designed for WIPER System

SWARMFEST 2007 17

Future Work Implement more crisis scenarios Study of aggregate patterns More realistic agent paths

Road and Highway use Vary Speed Traffic Jam Scenarios

Water Social network structure

SWARMFEST 2007 18

Questions?