Phoenix March 26 2009

(Converted from Microsoft PowerPoint Presentation, approximate slide layout retained)

Simulation of Emergency Evacuations using TRANSIMS

Dr.-Ing. Hubert Ley

Transportation Research and Analysis Computing Center Argonne National Laboratory

March 26, 2009

TRACC HPC Configuration Diagram

The TRACC computational cluster is a customized LS-1 system from Linux Networx consisting of 512 core 128 compute nodes, each with two dual-core AMD 2216 Opteron CPUs and 4GB of RAM, a DataDirect Networks storage system consisting of 240TB of shared RAID storage, expandable to 750TB, a high-bandwidth, low-latency InfiniBand network for computations, and a high-bandwidth Gigabit Ethernet management network. The system will also include the highest-performance compiler and MPI library available for the AMD Opteron architecture. with a peak performance of ~2 TFlops

Emergency Evacuations of the Chicago Business District

■ This project has been implemented to model the effects of a no-notice event on the multi-modal regional transportation system in the Chicago metropolitan area

■ The chosen scenario postulates a radioactive release following an explosion at the base of the Sears Tower

■ This project deals with the dynamic effect on the transportation system

Simulated image of radioactive release in Chicago

Fundamental Capabilities of the TRANSIMS Approach

■ Multi-modal transportation (vehicles, buses, trains, walking, bicycles,...)

■ Extremely large simulation areas, e.g. Chicago (10,000 square miles)

■ Fully time-aware routing of each individual traveler for all travel modes

■ Microsimulation for large metropolitan areas to determine the interactions between travelers and vehicles to determine second by second movements

- Determination of vehicle interactions, such as lane changes, speed changes, passenger loading and unloading, .

- Interaction with the road network, e.g. with traffic signals, speed limits, turn lanes, transit vehicles, .

■ This approach overcomes the limitations of traditional traffic forecasting models:

- Delivering transportation system performance as a full function of time instead of static solution for a few time periods (e.g. am and pm peaks)

- Microscopic interaction between vehicles and travelers delivers accurate results compared to simple volume delay functions and aggregate data.

■ Main challenges: Massive demands on computation time and a need for extremely detailed input data

10,000 Square Miles Simulation Area

Available Data Sources and the Types of Data Needed

Overhead image of cars in the streets in Chicago with schematic of traffic flow

■ CMAP and TRACC Network Improvements

■ Main focus is on network topology, in particular:

- Connectivity

- Number of Lanes

- Functional Classes

- Speed Limits

- Coded Length

- Capacities, etc.

■ For visualization and more precise modeling:

- Exact geographic locations

- Shapes along links

- Correct integration of transit links and stops, etc.

Current

Status

■ Each

individual lane is modeled

■ Pocket

lanes are modeled

■ Lane Connectivity

■ Signals

- Phasing

- Timing

■ Parking

■ Many more details

Overhead view of Chicago with green points overlaid

The Regional Road Network

■ ~10,000 square miles

■ Road network

- 40,000 links

- 14,000 intersections

- 250,000 locations

■ ~28 million vehicle trips

■ ~1.5 million transit trips

■ Trip tables

- Break-down by purpose (work, truck, airport, and many more)

Overhead image of roadways (shown by yellow lines)

Overhead image of Chicago showing major roadways in red and orange

Some Preliminary Results

■ The metropolitan road network accommodates the trips reasonably well problems)

■ Traffic volumes per lane are shown as an indicator of congestion (e.g. 8:00 to 8:15)

■ The TRACC cluster has reduced computing time for 27 million routes to less than 15 minutes using just 48 processors (of 512)

Example Case Study: Evacuation with Transit to a Shelter

■ Scenario:

- The Emergency Response Team secures a few suitable transit stations

- People in the Evacuation Area are directed to walk to these stations

- Transit transportation is provided to take them to a shelter location (e.g. United Center)

■ TRANSIMS is already able to simulate for each individual person:

- Delays in leaving buildings

- Walking towards the closest evacuation stop

- Flow of buses within congested roads and/or on reserved lanes

- And more ...

Overhead image of Chicago showing designated areas of traffic flow in red, blue and green

Complex Evacuation Strategies

Scenario 4: Evacuation via Transit to Shelters, overhead view of Chicago

TRACC Contact Information

■ Director's Office

- Dave Weber, Director dpweber@anl.gov

- Mike Boxberger boxberger@anl.gov

■ Systems Administration

- Jon Bernard bernard@anl.gov

■ Traffic Simulation

- Hubert Ley hley@anl.gov

■ Computational Structural Mechanics

- Ronald Kulak kulak@anl.gov

■ Computational Fluid Dynamics

- Tanju Sofu tsofu@anl.gov



		*Simulation of Emergency Evacuations using TRANSIMS*
		*Dr.-Ing. Hubert Ley* Transportation Research and Analysis Computing Center Argonne National Laboratory March 26, 2009



	*TRACC HPC Configuration Diagram* The TRACC computational cluster is a customized LS-1 system from Linux Networx consisting of 512 core 128 compute nodes, each with two dual-core AMD 2216 Opteron CPUs and 4GB of RAM, a DataDirect Networks storage system consisting of 240TB of shared RAID storage, expandable to 750TB, a high-bandwidth, low-latency InfiniBand network for computations, and a high-bandwidth Gigabit Ethernet management network. The system will also include the highest-performance compiler and MPI library available for the AMD Opteron architecture. with a peak performance of ~2 TFlops



	*Current* *Status* ■ Each individual lane is modeled ■ Pocket lanes are modeled ■ Lane Connectivity ■ Signals - Phasing - Timing ■ Parking ■ Many more details



	*The Regional Road Network* ■ ~10,000 square miles ■ Road network - 40,000 links - 14,000 intersections - 250,000 locations ■ ~28 million vehicle trips ■ ~1.5 million transit trips ■ Trip tables - Break-down by purpose (work, truck, airport, and many more)



		*Some* *Preliminary Results* ■ The metropolitan road network accommodates the trips reasonably well problems) ■ Traffic volumes per lane are shown as an indicator of congestion (e.g. 8:00 to 8:15) ■ The TRACC cluster has reduced computing time for 27 million routes to less than 15 minutes using just 48 processors (of 512)



	*Example Case Study: Evacuation with Transit to a Shelter*

	■ Scenario: - The Emergency Response Team secures a few suitable transit stations - People in the Evacuation Area are directed to walk to these stations - Transit transportation is provided to take them to a shelter location (e.g. United Center)

	■ TRANSIMS is already able to simulate for each individual person: - Delays in leaving buildings - Walking towards the closest evacuation stop - Flow of buses within congested roads and/or on reserved lanes - And more ...



	*TRACC Contact Information*

	■ Director's Office - Dave Weber, Director dpweber@anl.gov - Mike Boxberger boxberger@anl.gov ■ Systems Administration - Jon Bernard bernard@anl.gov ■ Traffic Simulation - Hubert Ley hley@anl.gov ■ Computational Structural Mechanics - Ronald Kulak kulak@anl.gov ■ Computational Fluid Dynamics - Tanju Sofu tsofu@anl.gov