Comparing Dynamic Traffic Assignment Approaches for Planning Ramachandran Balakrishna Daniel Morgan...

Comparing Dynamic Traffic Assignment Approaches for

Planning

Ramachandran BalakrishnaDaniel Morgan

Qi Yang

Caliper Corporation

12th TRB National Transportation Planning Applications Conference, Houston, Texas

20th May, 2009

Outline

• Introduction• Motivation• DTA comparison methodology• TransModeler 2.0 overview• DTA in TransModeler 2.0• Empirical tests• Conclusion

Introduction• Within-day dynamics: I-405, Orange County, CA

0

2000

4000

6000

8000

10000

12000

14000

0 50 100 150 200 250 300 350

Time of Day

Flo

w (

veh

/ho

ur)

Hourly Flows

5-Min Flows

[Source: PeMS on-line database]

• Temporal variability– Complex interactions of network demand– Aggregation error

Introduction (contd.)

• Static traffic assignment– Cannot capture detailed within-day dynamics– Does not handle capacity constraints, queues

• Produces unrealistic results (e.g. flow >> capacity)

• Dynamic traffic assignment (DTA)– Models temporal demand, supply variations

• Uses short time intervals, usually 5-30 minutes

– Captures capacity constraints, queues, spillbacks

– Superior to static for short-term planning• Evacuation & work zone planning, dynamic tolls,

etc.

Motivation• Different types of DTA

− Analytical− Simulation-based (micro, macro, meso)

• Tradeoffs perceived between realism, running time

• Methods are often chosen based on available computing resources

• Objective comparison of different methods is lacking

DTA Comparison Methodology

• Objective: User equilibrium (UE)− Dynamic extension of Wardrop’s principle− Same impedance (e.g. travel time) for all

used paths between each OD pair, for a given departure time interval

• Test DTAs on common platform and dataset

• Measure and compare convergence− Relative gap− Convergence rate

TransModeler 2.0 Overview

• Simulates urban traffic at many fidelities− Microscopic (car following, lane changing)− Mesoscopic (speed-density relationships)− Macroscopic (volume-delay functions)− Hybrid (all of the above)

• Employs realistic route choice models• Handles variety of network

infrastructure− Signals, variable message signs, sensors,

etc.

• Simulates multi-modal, multiple user classes

DTA in TransModeler 2.0

• Analytical (Planner’s DTA)− Based on Janson (1991), Janson & Robles

(1995)

• Simulation-based DTA− Feedback approach− Iterates on simulation output until

convergence

• All DTAs are run on same network

DTA in TransModeler 2.0 (contd.)

• Simulation-based DTA

− Feedback methods• Path flow feedback• Link travel time feedback

− Fidelity• Microscopic• Mesoscopic• Macroscopic• Hybrid

Simulation-Based DTA Framework

• Path flow averaging

Simulation-Based DTA Framework (contd.)• Link travel time averaging

• Averaging method

• Choice of averaging factor− Method of Successive Averages (MSA)− Polyak− Fixed-factor

Simulation-Based DTA Framework (contd.)

Empirical Tests

• Columbus, Indiana− 6630 nodes− 8811 links− 85 zones

− AM peak period• 7:00-9:00• ~42,000 trips

Empirical Tests (contd.)

• Static assignment

− Relative gap• 50 iters: ~0.008• 100 iters: ~0.006 • 2000 iters:

~0.0005

− Run time• 50 iters: ~36 sec• 100 iters: ~1 min• 2000 iters: ~24

min

Empirical Tests (contd.)

• DTA− Feedback method: MSA

• Path flow averaging• Link travel time averaging

− Model fidelity• Microscopic• Mesoscopic

• Four experiments

Empirical Tests (contd.)

• Microscopic DTA results

Empirical Tests (contd.)

• Mesoscopic DTA results

Empirical Tests (contd.)

• Feedback with path flows

Empirical Tests (contd.)

• Feedback with link travel times

Conclusion

• Static assignment is fast with known properties, but does not capture dynamics

• Simulation-based DTA is more realistic but slower and harder to analyze

• Travel time feedback appears to be faster than path flow averaging for simulation-based DTA

• Tests on more networks are required

Analytical DTA Framework

• Planner’s DTA− Based on Janson (1991), Janson & Robles

(1995)

− Bi-level, constrained optimization • Outer: consistent node arrival times• Inner: User equilibrium for given node arrival

times

− Extended by Caliper:• Spillback calculations• Stochastic user equilibrium• Better travel times

− Reasonable results on large planning networks