Michael Mahut, Michael Florian & Daniel Florian

1. A new adaptive, multiscale traffic simulation
Michael Mahut | michaelm@inrosoftware.com
Michael Florian, Daniel Florian
AITPM, July 26-29 – Sydney, Australia

2. Motivation

3. Current Trends
• Larger-scale traffic simulation models are becoming
increasing popular
 Meso and micro simulation-based dynamic traffic assignment
 Metropolitan to regional scale
• Motivated by need for higher realism/fidelity
 Temporal dynamics/resolution
 More detailed network models
• Facilitates modeling of complex mechanisms
 Departure time choice
 Real-time pricing, e.g. congestion-based tolls
 Freeway management, traveller response to information, etc…

4. Challenges
• Well specified scenarios:
 Algorithms are very good at finding stable, converged solutions
 Existence and uniqueness of solution are observed empirically
• Unbalanced network demand and supply:
 Models become unstable: nonlinear response of network delay
to demand
 Gridlock conditions
 O-D impedances: no longer respond to demand changes
 Non-convergent, unusable solutions
• Conclusion: algorithms are not sufficiently robust

5. Example: highly congested scenario

6. High demand scenario

7. Why don’t we see this effect in static models?
• In a static model (vdf)
 Path travel time = function of link (+turn) v/c ratios on the path
• In a simulation model
 Path travel time is affected by bottlenecks that are not on the
path
 This is due to congestion spillback
• We will refer to delay from bottlenecks off the path as
secondary or indirect delay
 This is the component of delay which grows in a highly
nonlinear way when demand >> supply

8. A new approach - adaptive simulation
• Adaptive driver response to extreme congestion
 Gradual formation of emergent lanes utilizes spare turn
capacity
• Adaptation is triggered by a delay threshold
 Exogenous parameter, can be calibrated
• Impact on delay propagation
 Delay in emergent lanes is not propagated upstream
+ 5 min

9. Additional properties
• Adaptive mechanism reduces secondary delay only
 Does not affect primary delay
• Bottleneck (turn) capacities are easily respected
 Extremely low flows can increase to become very low flows
• No vehicles removed from the simulation
• No vehicles lose their paths
Outputs
• Length of individual emergent queues (in vehicles)
• Delay in individual emergent queues

10. Example 1: High Demand
No adaptation With adaptation
Impact of adaptive mechanism on model convergence

11. Example 1: High Demand
Length of emergent queues aggregated to nodes

12. Emergent queues through model iterations
5 iterations 10 iterations 20 iterations

13. Example 2: Very High Demand
No adaptation With adaptation
Impact of adaptive mechanism on model convergence

14. Example 2: Very High Demand
Length of emergent queues aggregated to nodes

15. Conclusions
• Fast models are not enough, we need
more stable models
• We propose an adaptive simulation
approach which responds to extreme
congestion
 Can be tailored to the degree of
congestion inherent in the scenario
 Ensures model stability even when
demand significantly exceeds supply
 Addresses scalability for larger
networks in congested conditions
 Provides a single traffic model at a
consistent level of detail over the entire
network

16. A new adaptive, multiscale traffic simulation
Michael Mahut | michaelm@inrosoftware.com
Michael Florian, Daniel Florian
AITPM, July 26-29 – Sydney, Australia