Calibrating a Region-wide

Microsimulation Model: Maricopa Association of Governments

TRB Innovations in

Travel Modeling 2014

Baltimore , MD

Daniel Morgan (daniel@caliper.com)

David Gerstle (david@caliper.com)

Caliper Corporation

Purpose

• Add analysis of operations and traffic simulation modeling to

the services MAG offers to its member agencies

• Build a model to complement MAG’s regional travel demand

model that:

• Has the operational sensitivity to capture effects of signal

operations, ITS projects

• Is able to capture the mobility benefits of major projects whose

impacts will be felt throughout Central Phoenix

• Accurately portrays the traffic impacts of transit improvements,

namely on high-capacity transit corridors

• Provides a calibrated base model from which smaller, more

focused studies can be derived

Approach

• Study Design Stage

• Solicit stakeholder input/support

• Plan model framework, design parameters, and geographic scope

• Model Data Preparation

• Assemble traffic count and signal timing data

• Develop simulation model network

• Framework Development and Testing

• Test, calibrate, and validate the model

• Training

Approach

• Study Design Stage

• Solicit stakeholder input/support

• Plan model framework, design parameters, and geographic scope

• Model Data Preparation

• Assemble traffic count and signal timing data

• Develop simulation model network

• Framework Development and Testing

• Test, calibrate, and validate the model

• Training

Design

• A model congruous with the regional travel demand model

• Objective: To achieve a degree of integration with the regional travel demand model such that they can share key model data seamlessly

• Solution: A simulation model in TransModeler capable of reading all file formats and data structures of the regional model in TransCAD and sharing a common zonal system (and, hence, ready exchange of origin-destination matrices)

• A multi-resolution traffic simulation model

• Objective: A simulation model with an appropriate balance of high-fidelity treatment of traffic flow phenomena and practical computational performance

• Solution: A microsimulation model enabling selective application of lower-resolution (e.g., meso) and multi-resolution (e.g., hybrid micro-meso) models

Development

1. Preparation of highly

detailed lane-level

geography/geometry

2. Import of centroids

and connectors from

regional model

3. Auto-adjustment of

TAZ connectivity

4. Manual addition of

centroids along study

area boundary

Development: Geography/Geometry

Development: Centroids/Connectors

Development: Centroids/Connectors

Calibration

1: Produce Initial Estimate

of O-D Traffic Demand

from Regional Travel

Demand Model

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

1

Calibration

2: Simulation-based

Dynamic Traffic

Assignment to Equilibrate

Route Choices

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

2

Calibration

3: Compare 15-min.

Simulated Volumes with

15-min. Segment and

Turning Counts

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

3

Calibration

4: Simulation-based

Dynamic O-D Estimation

to Improve Match with

Counts

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

4

Calibration

Iterate Steps 2-4: Repeat DTA

to “recalibrate” route choice

to the changes in demand

resultant from the ODME

step

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

4

2

3

Validation: 15-minute INRIX Speeds

Visual comparison model

speeds with INRIX speeds to

ensure start, severity, duration

of bottlenecks

Targeted adjustment of trip

table to improve match with

bottlenecks while maintaining

goodness-of-fit with counts

Calibration

ODME (Step 4) later

extended to incorporate

observed speeds

simultaneously with counts

Travel Model

Subarea

Analysis

Seed

Matrix

Traffic

Counts

Simulation-

based DTA

Travel

Times &

Delays

Simulation-

based ODME

Good

Fit? Finished

Simulated

Traffic

Volumes

No Yes

4 15-min.

INRIX

Speeds

Visual Audit

Do route choices comport with expectations, local knowledge?

Visual Audit

Do route choices comport with expectations, local knowledge?

Query paths traversing critical link, turning

movement, or arbitrary link sequence

Query used paths by

origin and destination

Goodness of Fit: RMSE

How well does the model match observed data?

AM Peak

0600-0659 0700-0759 0800-0859 0600-0859

Collector (N=62) 93.44% 73.86% 77.52% 76.75%

Arterial (N=1,162) 52.31% 40.18% 39.39% 39.36%

Ramp (N=49) 48.81% 45.98% 45.56% 45.07%

Freeway/Expressway

(N=220) 14.06% 14.28% 14.45% 12.85%

All (N=1,493) 33.03% 29.37% 29.17% 27.70%

PM Peak

1500-1559 1600-1659 1700-1759 1500-1759

Collector (N=62) 64.79% 62.78% 70.12% 62.76%

Arterial (N=1,162) 43.44% 33.25% 37.98% 35.51%

Ramp (N=49) 41.21% 41.03% 42.48% 38.73%

Freeway/Expressway

(N=220) 14.52% 15.03% 15.89% 13.70%

All (N=1,493) 30.73% 27.06% 29.80% 26.86%

Goodness of Fit: AM Scatter Plots

How well does the model match AM counts?

y = 0.9115x - 54.279R² = 0.9475

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ulat

ed C

ount

s

Field Observed Counts

y = 0.9302x - 45.489R² = 0.9375

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ula

ted

Co

un

ts

Field Observed Counts

y = 0.9254x - 36.982R² = 0.9419

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ula

ted

Co

un

ts

Field Observed Counts

y = 0.9383x - 221.49R² = 0.9511

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ula

ted

Co

un

ts

Field Observed Counts

Goodness of Fit: PM Scatter Plots

How well does the model match PM counts?

y = 0.924x - 46.772R² = 0.939

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ula

ted

Co

un

ts

Field Observed Counts

y = 0.9175x - 73.142R² = 0.9249

0

2000

4000

6000

8000

10000

12000

0 2000 4000 6000 8000 10000 12000

Sim

ula

ted

Co

un

ts

Field Observed Counts

Goodness of Fit: Spatial Pattern of Fit

What is the spatial distribution of the percent errors?

AM PM

Validation: Bottleneck Matching

How well

does the

model

reflect

critical

bottlenecks

in the study

area?

Applications

• US-60/Grand Avenue COMPASS Study

• Old Town Peoria Traffic Study

• Various analyses of traffic interchange

redesigns and other roadway

improvements