Simulationbased Dynamic Equilibrium

1
Simulation-based Dynamic Equilibrium
EMME/2 Users Group Meeting Mexico City, October
2004
2
Motivation

• There is a growing need for models that are
  dynamic, and that are more detailed than static
  models
• congestion is increasing and is not captured
  realistically by static models
• network management is becoming more
  sophisticated
• HOV lanes, bus preemption of traffic signals
• time-varying tolls
• VMS and other driver information systems

3
Current Solutions (1)

• Dynamic traffic models (macroscopic and
  microscopic) have been around for a long time,
  without becoming as widely used as static models
• dynamic models are more realistic, but as a
  consequence are not nearly as forgiving
• errors in network coding, O-D demands, and
  routing assumptions can cause huge problems with
  results, often due to excessive congestion

4
Current Solutions (2)

• Dynamic models accurately capture congestion
  spillback, from one link to another upstream
• demand/capacity errors on one link can spill back
  over many upstream links, cutting off flow
  through an entire area, even causing gridlock
• these errors are particularly sensitive to
• O-D demands (input)
• Routing of these demands (model)

5
Dynameqs Solution (1)
Network definition
Time-dependent OD matrices
Traffic Control Data
choose initial paths
calculate path flows (t)
run traffic simulation
determine path travel times (t)
STOP
convergence achieved?
NO
YES
6
Dynameqs Solution (2)

• Dynameq converges towards an equilibrium
  solution, but early iterations can exhibit very
  heavy congestion and lead to gridlock
• Dynameq uses a gridlock prevention algorithm that
  keeps traffic flowing
• this ensures that useful travel times are
  obtained at each iteration of the traffic
  simulator

7
Dynameqs Traffic Simulator

• Based on the three standard components
• car-following, lane changing, gap-acceptance
• These components are less complex than in other
  commercial simulation models
• some reduction in realism
• huge benefits in computation time, which is
  critical in an iterative method
• network much easier to code
• model much easier to calibrate

8
Where does it fit?
9
DEMO
10
(No Transcript)
11
PERFORMANCE

• Southern Calgary
• 750 links, 300 nodes, 79 zones, 12,500 cars
• 6 (10 min) departure intervals x 10 paths
• 25 Mb RAM, 30 iterations x 6.4 s 4.3 min
• Downtown Calgary
• 563 links, 406 nodes, 96 zones, 32,667 cars
• 6 (20 min) departure intervals x 10 paths
• 33 Mb RAM, 30 iterations x 15.5 s 10.3 min
• Stockholm
• 2080 links, 1191 nodes, 114 zones, 108,000 cars
• 8 (10 min) departure intervals x 10 paths
• 80 Mb RAM, 40 iterations x 1.5 min 1.0 hour

12
SOUTHERN CALGARY NETWORK
13
(No Transcript)
14
(No Transcript)
15
(No Transcript)
16
(No Transcript)
17
DOWNTOWN CALGARY NETWORK
18
(No Transcript)