A Dynamic Traffic Simulation Model on Planning Networks

Qi Yang Caliper Corporation

TRB Planning Application ConferenceHouston, May 20, 2009

Outline

• Motivation• Model structure• Input• Output• Case study• Next step

Motivation: An engine for DTA

• Static traffic assignment failed to capture the temporal dimension of traffic flows

• Time variant travel times (links and paths between OD pairs)– Estimation of congestion– Travel time skimming for activity based

models– Dynamic ODME

• Various DTA models available, and we need one which works in TransCAD

Requirements

• The need to represent:– Queues, shockwaves and spillbacks– Delay at intersections and bottlenecks– Traffic signal controls at intersections

• Why not microscopic traffic simulation?– Data is often not adequate to calibrate the

model– Computational requirement is extensive,

especially for large networks– Modelers need a cheap and fast solution

because of time and budget constraints

Proposed Model:Transportation Dynamic Network Analyzer

• TransDNA is a procedure which runs as a thread in TransCAD

• A path-based traffic simulator for moving individual vehicles between OD pairs

• “Completely” compatible with existing planning networks

• Reuse (begin from) the trip matrices in planning models

• Produce time-dependent travel times by links, paths, and trips

• Complementary to traditional 4-step model and a tool for new activity based models

Model Structure: Work Flow

4-StepPlanning Model

Turn Movement Counts

TransCAD MMA

Link Travel Times

IntersectionTraffic Control Plans

Signal Timing

Time-Variant Matrices

TransDNATraffic Simulation

Speed &Travel Times

Link and Turn Movement Counts

SeedOD Matrices

Dynamic ODME

Path Choice Model

DynamicMap Themes

Path TablesTrip Tables Capacities

Model Structure: Network Representation

EB 2

WB 3WB 2WB 1

EB 1

Link Segmentation

• Travel lanes• Added lanes on left and/or right• Movements allowed and lane grouping

Traffic Models

• Delay at intersections (global or node specific)– Signalized– Unsignalized

• Vehicle movements in links modeled by:– Speed/Density, or– Volume/Delay

Traffic Dynamics in Mesoscopic Simulation

In Real-world

In TransDNA

Van Aerde Model

21 3

1

f

kc

c c uu u

U

Uf

Uc

qc q

21 (2 ) /f c f j cc u u u k u

2 22 ( ) /f f c j cc u u u k u

23 1/ /c f j cc q u k u

Where:Greenshields Model

Pipes Model

Capacity Speed

Free Flow Speed

Capacity Jam Density

k and q has linear relationship

k and u has linear relationship

c1 = c3 = 0

(1) If uc = ½ uf , kc = ½ kj

(2) If uc = uf , c1 = 1 / kj

c2 = 0 , c3 = 1 / qc – 1 / kj uf

Source: Hesham Rakha and Brent Crowther

Input

• Network– Road classification (capacity, free flow speed,

etc)– Number of lanes and their length– Travel time variability

• Travel time tables– Historical– Updated

• Time-dependent OD matrices– By access control (HOV, trucks, etc)– By value of time (tolls and HOT)

• Intersection Signal Controls– Green splits– Delay by movements– Saturation flows by lane groups

• Model parameters

Output

• Trip table– Ori., Des., Path– Dep. Time, Arr. Time– Mileage, Delay

• Link passage– Vehicle ID, Time

Enter/Leave

• Link statistics– Vehicle Count, Speed,

Entry Queue

• Movement counts and delay

Case Study: I-270 Corridor, MD

• Subarea from PG/WashCOG– 2371 links and

928 connectors– 100,688 ODs

• Simulation– 6-9:00 AM peak– 571,000 trips– Runs 10-15 times

faster than real time w/ data recording on an i7 desktop (8 cores)

I-270/I-495 – Density

6:30 AM 8:30 AM

9:00 AM

Case Studies – Columbus, IN

• Full Planning Model– 8811 links and 984 connectors

– 7225 OD pairs (85x85)

• Simulation Result– 8-10 AM peak– 824,000 trips (not much congestion)

– Runs 40-45 times faster than real-time on i7 desktop (8 cores)

Link labels - where are the vehicles?

Next Step

• Complete the DTA and ODME loop• Model calibration & validation

based on field data• Support user defined SD and VD

functions• Testing and more testing … …

Thank You!

Van Aerde Model