Dynamic Traffic Assignment With Cube Avenue

Dynamic Traffic Assignment

With Cube Avenue

Colby Brown, PTP

Regional Director, USA and Canada

Macroscopic and

Static Modelling

Mesoscopic and

Dynamic Modelling

Microscopic and

Dynamic Modelling

Transportation modelling tools

A Comprehensive Traffic Modeling Suite

Cube Voyager for regional flows

Cube Dynasim for

corridor/subarea operations

Cube Avenue for dynamic traffic

assignment and bottleneck simulation

• Avenue is an optional add-on to

Cube Voyager that enables

Dynamic Traffic Assignment

with Mesoscopic Simulation:

• Dynamic Traffic Assignment

Routes and flow rates change during

the model period based upon

congestion

• Mesoscopic Simulation

Packets of vehicles are simulated as

they move through the network

What is Cube Avenue?

• Macroscopic models evaluate

congestion using speed-flow

curves – based on fundamental

traffic engineering relationships

• Microscopic models simulate

individual vehicle trajectories on

a detailed network and use

behavior models (e.g. car

following, gap acceptance) to

predict second-by-second driver

responses to en-route events

Congestion: the key to model scale

• Typically simulate movement of

trips along their routes at some

detailed resolution (packet)

• Discretely model traffic queues

in network at bottlenecks

(e.g. intersections, ramps, tolls)

• Traffic stream performance is

typically still evaluated using

aggregate macroscopic

(e.g. speed-flow) relationships

Mesoscopic Traffic Modelling

As Cube Avenue

processes the

various kinds of

information used

in dynamic traffic

assignment, it

moves between

different levels of

aggregation.

The temporal detail cycle

 The model duration is explicitly defined and divided into smaller

time segments

Simulation Time

Warm-up Period Model Period

Segment Segment Segment Segment

Post-load Period

Segment Segment

Demand

Time Segments

Demand Profile

The demand is assumed to be constant

during each time segment

Cube Avenue – Dynamic Traffic Assignment (DTA)

Method of Successive Averages

1.2 1.2

Path 1

0.6 0.6

Path 1

0.6 0.6

Path 2

0.4 0.4

Path 1

0.4 0.4

Path 2

0.4 0.4

Path 3

0.3 0.3

Path 1

0.3 0.3

Path 2

0.3 0.3

Path 3

0.3 0.3

Path 4

“Perfect” solution of MSA…

But maybe not practical… each iteration more packets…

Eventually you WILL run out of memory.

Packet Allocation Methodology

1.2 1.2

Path 1

1.2

Path 1

1.2

Path 2

1.2

Path 1

Path 2

1.2

Path 3

With each iteration, each packet has a MSA probability (1/Iter)

of being “Selected” for the fancy new path.

Globally the proper amount of trips are moved to the new path…

never runs out of memory or slows down.

Path 1

Path 2

Path 3

1.2

Path 4

1.2

• In extremely congested

networks, an “All or Nothing”

first iteration can cause

gridlock.

• Later time segments have no

hope of „learning‟ how to

navigate the network when

earlier time segments continue

to see drastic changes to their

routing.

• When later iterations finally

have a stable initial condition,

the amount of traffic being

adjusted is too small to be

meaningful.

• A time segment must wait its

turn to begin being assigned.

Loading in Congested Networks

Incremental Time Segment

• The user defines a number of iterations

that each time segment gets to itself

before the next time segment may start.

• Each time segment respects its own

convergence and maximum number of

iterations (maxiters)

Results

Converged

COMBINE=AVE, PACKETS=PA,

ITERLOADINC=0 (defaults)

• PACKETS

• PA (Packet Allocation)

• PS (Packet Splitting)

• ITERLOADINC

• # specifies the number of iterations to wait

before incrementally adding a time segment.

Scripting Dynamic User Equilibrium

(DUE)

• PARAMETERS

• MODELPERIOD= duration, in minutes

• SEGMENTS = list of durations, in minutes (sum >=

MODELPERIOD)

• VEHPERDIST = vehicles per lane per unit distance (fleet jam

density)

• LINKREAD keywords

• C = maximum flow rate capacity (vehicles per model period

duration)

• DYNAMIC C = list of segment-specific flow capacities, same units

as C

• STORAGE = link-specific storage value (overrides VEHPERDIST)

• ILOOP keywords

• DYNAMICLOAD instead of PATHLOAD, may use __TS__ in

expressions

• ADJUST keywords

• May condition LW calculations on built-in TIMESEGMENT variable

Scripting – Other Features

 Time-varying O-D travel demand

(flow per time segment)

 Link properties

(Capacity, T0, Storage, Junction Modelling, ...)

Cube Avenue – Input Data

Output Data: Dynamic Bandwidths

Output Data: Dynamic Intersection LOS

Output Data: Dynamic Queues

Output Data: Packet Log Animation in

GIS

The *.LOG text file

<!-- RUN="NETI.101585" PGM="AVENUE (v.10/18/2007

[4.2.0])" TIME="Thu Oct 18 18:47:46 2007"

NPKT=133997 PKTLNG=64 START=-0.500000

END=1.000000 NVOL=2 VPD=250.000000

TimeSliceEnd=-0.416667,-0.333333,-0.250000,

-0.166667,-

0.083333,0.000000,0.083333,0.166667,0.250000,0.33

3333,0.416667,0.500000,0.583333,0.666667,0.750000

,0.833333,0.916667,1.000000 -->

<v>

<ix=0,f=0.357143>

<ix=2,f=0.357143>

</v>

<r>

<n=20,a=*,d=-0.499593>

<n=430,a=-0.499593,d=-0.499529>

<n=447,a=-0.472814,d=-0.472814>

...

<n=475,a=-0.462416,d=-0.462416>

<n=464,a=-0.458362,d=-0.458362>

<n=476,a=-0.454521,d=-0.454521>

<n=478,a=-0.448999,d=-0.448999>

<n=484,a=-0.435480,d=-0.435480>

<n=486,a=-0.432114,d=-0.432114>

<n=17,a=-0.432114,d=*>

</r>

</p>

The text packet log follows a

pseudo-XML format that is left

intentionally open so as to permit

additional parsing, processing

and summary by the user.

Each file consists of a header

record followed by one or more

blocks of packet data.

Header

Record

Packet

Data

By applying record processing techniques to packet log output data,

you can implement many advanced analyses with Avenue:

• Build origin-destination table from log file

• Select node/link analysis

• Select link/node trip table: build table of trips using some node at some particular

time

• Check whether packets used a particular link/node and build a link table from the

list of nodes used by these packets (2 passes of the file)

• Extract average queue for specific packets (departure minus arrival)

• Temporal disaggregation (e.g. build 15-minute matrices from peak

hour simulation output based upon recorded departures)

• Peak spreading

• Build packet table from log file, flag packets that failed to arrive at their destination

• Shift packets to new departure time segment based upon logit or other decision

rule

• Re-build hourly trip matrix from packet table

• Other applications: ITS/VMS, parking, sub-area extraction, ME, etc…

Processing the Packet Log Data

DTA – OD Estimation