ACS Lite Project Overview - Kittelson & Associates, Inc.signalsystems.kittelson.com/ppt/ACSLITE TRB...

U.S. Department of TransportationFederal Highway AdministrationFHWA Contract No.DTFH61-02-C-00047

ACS Lite Project OverviewRaj Ghaman, P.E. , FHWA

Douglas Gettman, Ph.D., Siemens ITSSteve Shelby, Ph.D., Siemens ITS

Presenting at:

Transportation Research Board Annual MeetingAdaptive Traffic Signal Control Workshop

Washington, DCJanuary 11, 2004

ACS LiteTRB - Adaptive Traffic Signal Control WorkshopWashington, DCJanuary 11, 2004


Slide 2

Outline

ACS = Adaptive Control Software

• Project goals and status• What’s Lite about ACS Lite?• ACS-Lite system architecture• ACS-Lite algorithms overview• Performance results• Questions?



Slide 3

FHWA’s Motivation for the ACS-Lite Project

• Limited U.S. deployment of ACS– 8 agencies as of 1999

• FHWA ACS research– RHODES, OPAC, RTACL

• ACS survey & ITE roundtable– 70% say ACS too costly– 40% unconvinced of benefits over TOD/TRPS– ACS too sensitive/dependent on communications & detectors– Difficult to understand, configure, and maintain

• Closed-loop systems are prevalent in marketplace– Can we develop an adaptive solution augmenting existing

hardware?



Slide 4

FHWA’s ASC-Lite Project Goals

WIDELY DEPLOYABLE adaptive control– Low cost design– Leverage existing infrastructure

• Work with closed-loop systems & standard actuated controllers• Standard fully-actuated detector layouts• Communications bandwidth & protocols• Standard NTCIP interface• Field deployable without connection to TMC

– Meet performance expectations



Slide 5

Project Team

Raj GahmanGene McHaleFelipe Luyanda

Larry HeadDoug GettmanSteve Shelby

Darcy BullockNils Soyke

Pitu MirchandaniSanjay Sridhar



Slide 6

Project Partners

TSIS/CORSIM integration & FHWA TReL testing

Charlie Stallard

Controller / Closed-Loop Signal System Vendors

Mark Hudgins

Gary Duncan

Peter Ragsdale

Ed Bertha



Slide 7

Project Summary

• Started in March 2002– Siemens ITS, Purdue, Arizona– Upgrade CORSIM (ITT Industries) for NTCIP interface– Partnership with NEMA controller manufacturers

• Eagle, Econolite, McCain, PEEK– Focus on arterials in initial phase, networks at a later time

• Status– Initial software development complete– Initial simulation evaluation complete– Initial phase final report available March 2004

• Coming soon– Hardware-in-the-loop testing at Turner Fairbank Traffic Research Lab (TReL)– Field testing with participating NEMA systems– Additional R&D of algorithms and additional components



Slide 8

What’s “Lite” about ACS-Lite?

LiteLiteLiteInexpensive Compatible

Easy to use

Effective

“Low-Hanging Fruit”



Slide 9

Typical Closed-loop System

Master ControllerMaster Controller

Local ControllerLocal Controller Local ControllerLocal Controller Local ControllerLocal Controller

Central System Softwarein <50% of deployed

systems

9600 Baud Dial-up

1200 Baud, Twisted Pair ~50% ofsystems

9600 Baud ~50% of systems

Semi-Actuated Detection >50% of systemsFully-actuated Detection <50% of systems

Main Arterial

8-10 Signals Per SystemM

inor

Cro

ss S

treet

Min

or C

ross

Stre

et

Min

or C

ross

Stre

et

3-7 Timing Plans onTOD Schedule

Limited use of TRPS



Slide 10

System Architecture

Field Master

Central System

ACS-Lite

ACS-Lite API(NTCIP-protocol)

Laptop

Protocol Translation

Service

Vendor-specific protocol



Slide 11

Turner-Fairbank TReL Testing Configuration

Field Master

Central System

ACS-Lite


Laptop

CORSIMSimulation

CID

CID

CID

ControllerInterfaceDevice

Protocol Translation

Service



Slide 12

Current Laboratory Configuration

ACS-Lite


Laptop

CORSIMSimulation



Slide 13

ACS-Lite Algorithms Architecture

Time-of-Day Tuner

Run-time Refiner

TransitionManagement

Arterial

TOD Plans- cycle, splits, offsets- pattern switch times

Plan Changes- transition method

MasterController

local local local local

Active Plan- cycle, splits, offsets- active pattern



Slide 14


Time-of-Day Tuner

Run-time Refiner


Arterial



MasterController



Smooth, incrementalparameter changes

Large & fastparameter changes

Fast & non-disruptiveparameter changes



Slide 15

Run-Time Refiner

• Adjust active timing plan– Cycle (TBD), splits, offsets– Small, incremental adjustments (not permanent - TBD)– Switch earlier or later to next pattern (TBD)

• Monitor real-time status– Detector volume & occupancy

• Sample every few seconds for cyclic flow profiles• Sample during green, yellow, & red intervals for phase utilization

– Actual phase durations of actuated controller– Reasons for termination (max-out, gap-out, etc.)



Slide 16

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)

Illustration of Run-Time Refiner

AverageDaily

Demand



Slide 17

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)

Illustration of Run-Time RefinerMonth’s

High-LowRange



Slide 18

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


Mid-day Plan

Free

AM Plan PM Plan

Free



Slide 19

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


TOD PlanCapacity



Slide 20

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


Today’sDemand

This is a perfectly average day



Slide 21

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


Run-Time Refineradjusts active plan

Remove excess capacity of TOD plan



Slide 22

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


This is a perfectly “average” day



Slide 23

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


This is a perfectly “average” day

This is today’s demand



Slide 24

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


Better utilizes capacity of streets, but does not build new lanes

Run-Time Refiner reallocates time where

needed



Slide 25

Transition Manager

• Manage controllers’ transition from one plan to next– Select existing transition mode

• Dwell• Add• Subtract• Best way (of Add/Subtract)

– Command sequence of changes (TBD)

• Transition Objectives– Timely return to coordination– Minimally disruptive



Slide 26

Illustration of Transition Manager

A B A B A B A B

B A B A B A B A

AB

A B

Offset

Offset

Dwell Only

Long Way

Short Way

Plan 1

Plan 2

1

1 1 1 1

2 2 2 2

2

2

2

2



Slide 27

Time-of-Day Tuner

• Periodically re-tune Time-of-day (TOD) plans (TBD)– Adjust cycle, offset, & splits– Changes are “permanent”– Fine-tune schedule of pattern switch times

• Benefits– Avoid additional 3-5% delay/year due to changing traffic

patterns– Remain effective during controller comm. failure– Plans tailored to accommodate daily variability– Respond to seasonal changes in traffic conditions



Slide 28

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)

Illustration of Time-of-day Tuner

Mid-day Plan

Free

AM Plan PM Plan

Free



Slide 29

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


What if variability changes?



Slide 30

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


What if variability changes?



Slide 31

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)

Illustration of Time-of-day TunerTime-of-Day Tuner adjusts to

handle extremes better



Slide 32

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


What if total flow changes?



Slide 33

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


What if total flow changes?



Slide 34

0200400600

80010001200

12:00 2:00 4:00AM6:00 8:00 10:00 12:00 2:00 4:00 6:00 8:00 10:00

Time of DayPM

Flow(vphpl)


Adjust TOD plan

Start AM plan earlier

Permanently increase max greens

8001000Hold AM

plan later



Slide 35


Time-of-Day Tuner

Run-time Refiner


Arterial



MasterController



Second-by-Second

Cycle-by-Cycle

Day-by-Day

ACS-Lite is “Appropriately” Adaptivewithin the data/communications

environment of closed-loop systems



Slide 36

Run-Time Refiner Algorithm Details

• Splits• Offsets



Slide 37

21

ACS Lite Split Adjustment Guidelines

• “EQUISAT” is most popular adaptive split strategy• Volume & model parameters can be unreliable• Use phase timing & termination data (not alone)• Use lane independent green occupancy data• Account for early-return-to-green• Reduce stops with intelligently biased splits• Smart biasing requires arrival profile knowledge

CapacityProgression



Slide 38

Split Adjustment Algorithm

Extend EQUISAT concept to multi-ring controllers• STEP ONE: Form reasonable estimates of degree

of saturation• STEP TWO: Minimize the maximum level of

saturation on any phase– Ensure barrier alignment & cycle time constraints are

satisfied– Accommodate progression by allowing lower level of

saturation on coordinated phases



Slide 39

Detector Layout

Flow ProfileDetectors

Phase Utilization Detectors

Need detectors at stop-bar of coordinated phases



Slide 40

Multi-ring Controller Terminology

• Barrier group (or just group) – The set of all phases (or ring-groups) between two

barriers (or all phases if there are no barriers)– 2 groups below: {1,2,5,6} and {3,4,7,8}

• A ring group is the set of phases on a ring in a group– 4 ring-groups: {1,2}, {5,6},{3,4}, and {7,8}

8765b

43a

21b



Slide 41

Balance saturation within ring-group

• Less split time => higher saturation• More split time => lower saturation

(1) 65%

(1) 70%

(1) 80% (2) 80%

(2) 85%

(2) 90%

Original splits for phases (1) & (2)

Worse splits for phases (1) & (2)

Better splits for phases (1) & (2)

90%

85%

80%

Degree of saturation estimates for each split allocation

MAX

Duration of ring-group



Slide 42

Balance saturation across barrier groups

95%

90%

85%

+10-1

Degree of saturation estimates for each barrier group duration

Duration of barrier-group

(2) 85%(1) 80%

(2) 80%(1) 80%

(2) 80%(1) 75% (4) 95%(3) 70%

(4) 90%(3) 70%

(4) 85%(3) 70%


-1

0

+1



Slide 43

Accounting for all rings

95%

90%

85%

+10-1

Degree of saturation estimates for each barrier group duration


(2) 85%(1) 80%

(2) 80%(1) 80%

(2) 80%(1) 75% (4) 95%(3) 70%

(4) 90%(3) 70%

(4) 85%(3) 70%


-1

0

+1

Inf.

83%

87%

+10-1 (6) 87%(5) 80%

(6) 83%(5) 80%

(6) 80%(5) 80% Infeasible

(4) 80%(3) 60%

(4) 75%(3) 60%

-1

0

+1



Slide 44

Typical split adjustment profile

0

10

20

30

40

50

60

70

7:00 7:05 7:10 7:15 7:20 7:25 7:30 7:35 7:40 7:45 7:50 7:55 8:00 8:05 8:10 8:15

Simulation Time

Cum

ulat

ive

Split

Tim

e

Split 8

Split 7

Split 6

Split 5

0

10

20

30

40

50

60

70

7:00 7:05 7:10 7:15 7:20 7:25 7:30 7:35 7:40 7:45 7:50 7:55 8:00 8:05 8:10 8:15

Simulation Time

Cum

ulat

ive

Split

Tim

e

Split 4

Split 3

Split 2

Split 1



Slide 45

ACS Lite Offset Guidelines

• Measure cyclic flow profiles directly• Account for travel time from the detector to the signal• Account for variable start of green• Account for both coordinated approaches and effect on

downstream signals• Maximize the total amount of captured flow

– Two options:• On inbound and outbound movements at ALL signals on the arterial• On inbound and outbound movements at EACH signal on the arterial independently

• Make small incremental changes to minimize transitions



Slide 46

Detector Layout

Flow ProfileDetectors

Phase Utilization Detectors

Need detectors at stop-bar of coordinated phases



Slide 47

0

20

40

60

80

100

120

0 90 180

Cycle Counter (seconds)

Occ

uanc

y (%

/sec

)

Local Offset Tuning

Arrival Flow

Shift to capture most arriving flow

Green Window



Slide 48

Account for all coordinated approaches

0

20

40

60

80

100

120

0 90 180


Occ

uanc

y (%

/sec

)

0

20

40

60

80

100

120

0 90 180


Occ

uanc

y (%

/sec

)Southbound

Northbound



Slide 49

Account for all downstream signals

0

20

40

60

80

100

120

0 90 180


Occ

uanc

y (%

/sec

)Upstream

Shifting earlier reduces stops locally

0

20

40

60

80

100

120

0 90 180


Occ

uanc

y (%

/sec

)

Downstream

Upstream shift would increase stops



Slide 50

Typical offset adjustment profile

-60

-50

-40

-30

-20

-10

07:

00

7:05

7:10

7:15

7:20

7:25

7:30

7:35

7:40

7:45

7:50

7:55

8:00

8:05

8:10

Time (min)

Off

set (

s)

Colors represent different CORSIM

runs



Slide 51

Simulation Performance Testing

• ITT Industries– Developed NTCIP agent interface to CORSIM– Developed multi-pattern capability and realistic transition logic

• Purdue– Developed “real-world” test scenarios– Synchro-optimized timings– Many, many, many simulation runs and independent assessment of results



Slide 52


• Evaluate algorithms parameters– Re-adjustment intervals (5 to 10 minutes)– Offset changes and max deviations (2 to 20 seconds, “any”)– Split adjustments and max deviations (2 to 20 seconds, “any”)– Results tend towards shorter re-adjustment intervals and larger

flexibility of algorithm to make adjustments

• Start with optimized timings – can ACS-Lite improve?• Start with bad/arbitrary offsets or splits – can ACS-

Lite find a good solution?• Change turning proportions and volumes to represent

real-world traffic changes – can ACS-Lite adapt?



Slide 53


Changes in volumes at side-street approaches to intersections 2, 4, and 7

impact the entire network

Vehicle minutes

Vehicle minutes



Slide 54

Evaluation Results – Total Control Delay

0.0

2000.0

4000.0

6000.0

8000.0

10000.0

12000.0

14000.0

16000.0

18000.0B

-100

0

A-1

000

B-1

111

A-1

111

B-1

112-

A

A-1

112-

A

B-1

223

A-1

223

B-1

233

A-1

233

B-2

000

A-2

000

B-2

111

A-2

111

B-2

112-

A

A-2

112-

A

B-2

223

A-2

223

B-2

233

A-2

233

B-3

000

A-3

000

B-3

111

A-3

111

B-3

112-

A

A-3

112-

A

B-3

223

A-3

223

B-1

333

A-1

333

Scenario

Tota

l Del

ay (v

eh-m

in)

Baseline

ACS-Lite

SS↓SS↓SS↓SS↑

SS↓ SS↓SS↑

SS↓ SSSS

AM Scenarios PM Scenarios Off-peak Scenarios

~38% improvement

Bas

elin

e

Bas

elin

e

Bas

elin

e



Slide 55

Evaluation Results – Phase Failures

0

5

10

15

20

25

30

35

40

45

103-

1

101-

1

104-

1

2-1

105-

2

1-2

106-

2

3-2

107-

3

2-3

108-

3

4-3

109-

4

3-4

110-

4

5-4

111-

5

4-5

112-

5

6-5

113-

6

5-6

114-

6

7-6

115-

7

6-7

116-

7

102-

7

Approach Link

Num

ber o

f Fai

lure

s

Synchro Baseline

ACS-Lite

Worst Case

Total Number of Phase Failures:

Synchro Baseline: 17ACS-Lite: 24Worst Case: 88 ~70%

improvement



Slide 56

“Very High Altitude” Evaluation Results

ACS-Lite provides consistent delay reduction

Delay ( -38%, -7.4% )Travel Time ( -6.4%, +3.5% )

Changing volumes & turning proportions

ACS-Lite usually makes improvement

Delay (-3.3%, +2.2%)Travel Time (-4.9%,+6.8%)

Start with bad side-street Splits

(no offset adjustments and progression bias)

ACS-Lite can find a good set of offsets

Delay (-4.2%, +0.9%)Travel Time (-4.0%,+1.3%)

Start with bad Offsets(no split adjustment)

ACS-Lite “does no harm”Delay (-0.0%, +0.7%)

Travel Time (-0.6%,+2.4%)Start with optimized

settings

Conclusionvs. “Do nothing”, initially as ACS-Lite

ACS-Lite test scenario



Slide 57

Conclusions

• Core ACS-Lite development is complete– Run-Time Refiner– Transition Manager– Communications and algorithms software infrastructure

• Performance evaluation in simulation is encouraging• Current configuration designed for up to 12 intersections

on arterial• Coming up

– TReL testing with Hardware-in-Loop – Field testing– Time-of-day Tuner algorithms development



Slide 58

Questions?

Hunt us down for a demo

ACS Lite Project Overview - Kittelson & Associates, Inc.signalsystems.kittelson.com/ppt/ACSLITE TRB...

Documents

Transcript of ACS Lite Project Overview - Kittelson & Associates, Inc.signalsystems.kittelson.com/ppt/ACSLITE TRB...