Transit Signal Priority
Peter
P
t G.
G Furth
th Northeastern University
Transit Signal Priority –
Help or Hype?
• Zurich: nearly zero traffic delay for trams, even with mixed traffic (and punctuality!!)
• San Diego trolley: green wave through
downttown inttersecti
tions
• Some US app
pplications: < 3 s savinggs per
intersection , or …
• No measurement at all
all
2
Transit Priority as a Societal
Societal
Objective
Objective
• Transit use benefits society in terms of
of
– Congestion
– Air quality, climate impact, energy use
– Vibrant communities
• Priority breaks the vicious cycle in which congestion drives people to switch from
from
transit to car
3
Priiority
i Mak
kes Sense
• One extreme ($$$$): build a metro
• Other extreme: do nothing, buses become
swamped in congestion
– Traffic delay can represent 30% of a bus route
route’ss
operating cost
– Feeds vicious cyycle of ever-lower transit use
• In between:
– Priority in space: bus lanes
lanes, etc
etc.
– Priority in time: signal priority
4
What Priority Means to a Transit
Operation
• Reduced running time ($$, passenger benefit).
Measures:
– Delay per intersection
– Average route running time
• Improved reliability ($$, passenger benefit).
– Recovery time, cycle time based on 95-percentile
running time
– Tri-Met Line 12: reduced needed cycle length from 104
min to 93 min (11%)
5
Operational Control: Schedule Adherence, Crowding
Schedule deviation along the
route, without priority,
Eindhoven
With priority
6
Priority Makes Transit …
• More competitive
– Bus has natural disadvantages due to stops and
walkingg / waitingg
– Priority compensates, especially if cars suffer
congestion
• More socially acceptable (red carpet)
– my great aunt …
7
Intelligent Signal Priority
• Not “preemption” (too blunt)
• Not “cautious priority” (almost useless)
• Intelligent tactics,
tactics algorithms,
algorithms detection to
give transit near-zero-delay service, without
undue
d iimpact on othher traffic
ffi
8
Green Extension
• Built-in logic in modern controllers
• Large benefi
fit to a few buses
– Therefore little disruption to traffic
• Extension
E t i increment
i
t is
i a parameter
t (10 s?? 15 s?)
?)
With green extension
Without priority
c u m u la t iv e
v e h ic le s
c u m u la t iv e v e h ic le s
b u s d e la y
b u s d e la y
v
1
red
1
s
g reen
v
t im e
1
X
n o rm al
s
1
red
g reen
9 t im e
“Priority Push” is not the same as Allowed Extension
Without priority
With green extension
c u m u la t iv e
v e h ic le s
c u m u la t iv e
v e h ic le s
b u s d e la y
b u s d e la y
v
1
red
1
s
g reen
R2
1
E[delay ]
=
2C 1 −
v s
v
t im e
1
X
n o rm al
s
1
red
g reen
t im e
⎡ R X 2
2
⎡
⎤
⎤
R2
1
1
E[delay
]
=
−
⎢ X
−
(1 − v s)
⎥
2
C
1
−
v s
⎣
C
2
C
⎦
Priority
P
i i
push!
Priority Push vs. Green Extension
(cycle length = 100 s, red time = 50 s, degree of saturation = 85%)
12 0
12.
10. 0
8 0
8.
Push
6. 0
( s)
4 0
4.
2. 0
0 0
0.
5
10
15
Gr een Ext ensi on ( s)
20
Priority Push vs. Red Time for
allowed green extension of 15 s
( l length
(cycle
l th = 100 s, degree
d
off saturation
t ti = 85%)
16. 0
16
14. 0
12. 0
10. 0
Push
8. 0
( s)
6 0
6.
4. 0
2. 0
0. 0
30
50
70
Red Ti me ( s)
90
Detection
Sketch 1
• Check-in
h k i detector
d
location
i
– Early enough to allow time to respond
– Late enough to estimate bus arrival time
• Checkout detector to cancel request
– Avoid wasted green
– Performance measurement
•
•
•
•
In-gground vs. overhead
Optical signal with calibrated sensitivity
Co t uous detect
Continuous
detection
o (sshort-ra
o t ange
ge radio)
ad o)
Automatic vehicle location (AVL) system
13
Upstream
U
t
Detector,
D t t with
ith
travel time = maximum green extension
Simplicity: • Request = detection
• No need for “priority
request generator”
Weaknesses:
W
k
• assumes constant speed
• no fl
flexibili
ibility for upd
dates,
time of day settings
• may nott be
b suit
itable
bl for other
th
priority tactics
14
What if There’s a Near-Side Stop?
• Detector located just after stop
• Disable optical signal until door closes
(Portland OR)
(Portland,
• Does vehicle queue block entry to the bus stop?
15
Common Weaknesses in Signal
Signal
Priority Implementation
Implementation
1. Lack of checkout detector = wasted green
2. Is extra time “borrowed” or “stolen”?
– Lack of compensation creates large queuing impacts
3. “Cautious priority”
– Inhibit priority for 5 minutes or 1 full cycle after a priority
i
interruption
i
– Inhibit priority if cross street occupancy exceeds threshold
4 Lack
4.
L k off dat
d ta coll
llection
ti and
d analysis
l i
– Nobody ever gets it right the first time
16
Natural compensation with fully actuated (uncoordinated) operatiions
• Transit
it phase
– Longer green in cycle with priority
– Shorter green in next cycle, because some of its
demand was served in previous cycle
• Competiing ph
hases
– Longer red due to priority
– Need more green, and get it, in next realization
• System quickly recovers
17
Lack of compensation in coordinated systems •
•
•
•
Fixed cycle length
Fixed point = end of coordinated phase
Uncoordinated phases may run shorter than their allotted split, but not longer
Coordinated phase gets the slack time (starts early)
1
SBT SBL
2
3 NBL
4
EBT
WBL
WBT
WBL
EBL
EBT
EBL
NBL NBT
SBT
5
WBT
6
E-W Street
Suppose coordinated phase is 2 (EBT)
• Green extension for 2 – no mechanism for compensation
• Green extension for 4 – slight compensation
NBT
7
SBL
8
N-S Street
18
Returning to Coordination
In coordinated systems, recovery means
• dissipating queues AND
• Returning to the background cycle
– “Short way” = shorten phases following an
extension
– “Long way” = lengthen phases, skip a cycle
19
Early Green
Truncate and possibly skip preceding phases
• What’s the truncation rule?
– How much to shorten competing phases? Can they be skipped?
• Smaller benefit to large number of buses
– More traffic impact; hard to implement when bus frequency is high
20
Early Green Issues
• Exclusive lane for bus? (No queue, easier arrival time
prediction)
• Mixed traffic: Eindhoven’s “electronic bulldozer”
• Arrival time prediction
– How far upstream? (Bus stop and intersection spacing …)
– Tracking queue length to know how long is needed to flush out the
queue (Zurich’s
(Zurich s trap logic)
21
Early green (truncating competing
phases) under coordination
1
2
3 NBL
4
EBT
SBT
WBL
EBL
5
WBT
6
E-W Street
NBT
7
SBL
8
N-S Street
Assume
ssu e phase
ase 2 iss coo
coordinated
d ated
• Early green for 2 is possible, but without compensation to
shortened phases
• Early green for 3 – not possible under standard logic
22
Eindhoven
i dh
Experiment
i
+
11
+
Coordinated phase
(ring road, no buses)
12
38
+
27
25
+
31
36
2
3
32
9
8
+
39
+
+
23
34
6
+
Figure by MIT OpenCourseWare.
Buses every 10 min NB and SB
23
Existing Priority Scheme
• Coordinated phase is 6
• Buses are on 4 and 8
(every 10 minutes)
• Priorityy onlyyif bus is
more than 20 s late
(about half the buses)
1
2
3 NBL
4
EBT
SBT
WBL
EBL
5
WBT
6
E-W Street
NBT
7
SBL
8
N-S Street
• G
Green
ee extension
e te s o
• Aggressive early green: reduce intervening phases to
minimum green
• “Short way” minimum green to return to background cycle 24
Intersection Experiment and Site Description
O dday each
One
h off
Entry
• priority to all buses (absolute)
• priority to late buses (conditional)
Exit
Exit
•no priority
•no
Entry
Exit
Exit
Camera facing each approach
Film “trap” between entry & exit
Playback in lab
Entry
Entry
Experiment Eindhoven
35
30
25
20
15
10
5
0
No Priority
y
Abs Priority
AM
South
d
Bound
North
d
Bound
South
d
Bound
Cond priority
North
d
Bound
Delay
y Public
c Transp
port
Averag
ge Transit Delayy [sec]
[ ]
PM
Experiment Eindhoven: Traffic Impact
100
90
80
70
60
50
40
30
20
10
0
No Priority
y
Abs Priority
AM
PM
5-6
4-5
3-4
2-3
10-11
9-10
8-9
Cond Prioriy
7-8
Delay Tim e
Average Vehicle Delay [sec]
Experiment Eindhoven
Averag
ge Vehicle Delayy per
p App
pproach [[sec]]
140
No Priority
Abs Priority
100
Cond Priority
80
60
40
20
AM
PM
Sou
uth
Bou
und
Ea
ast
Bou
und
No
rth
und
Bou
We
est
Bou
und
Sou
uth
Bou
und
Ea
ast
Bou
und
No
rth
Bou
und
0
We
est
Bou
und
Delay Tim
me
120
Experiment Eindhoven
Relative Cap
pacity
yp
per App
pproach ((no priority
y= 100% )
120
N P
No
Priority
i it
100
Abs Priority
80
ond Priority
60
40
20
AM
PM
Westbound
Eastbound
Sout hbound
Nort hbound
Westbound
Eastbound
Sout hbound
Nort hbound
0
Lessons from Eindhoven
• Aggressiive early
l green result
ltedd in
i near-zero delay
d l
for buses
• Conditional priority needs finely-tuned schedule
– Schedule too tight – bus always late – absolute priority
– Schedule too loose – bus always early – no priority
• Lack of comp
pensation: OK for 6 interrupptions per
hour, but not 12
• Capacity loss due to
– Early green truncations, but more from …
– “short
–
short way”
way recovery to background cycle
30
Early Red
Shorten bus street’s current green to get faster return to green in thhe next cyclle
• Needs advanced detection (almost a full cyycle))
• Incompatible with typical coordination logic
– custom programming
31
Flush and Return
Flush-and-Return
Early green tactic for Near-Side Stops
Tested using simulation on San Juan (PR) arterial
• Green extension to clear queue from bus stop
• Force signal to red
during stop
– Minimizes bus’s impact on road capacity
• Return to green as
quicklyy as possible
32
Phase Rotation
• Example:
E ample: change leading bbuss phase to lagging
– Lagging bus phase becomes leading – like early green
– Leading
L di bus phhase becomes laggi
l ing – lik
like green extensiion
(more effective)
• U
Used
d exttensiivelly in Germany
• Zurich’s pre-application safety campaign: random
phase sequencing for 6 months!
33
Phase Insertion
• Secondd reali
lizatiion on bus detectiion onlly
• Greatly reduces red period for bus – big reduction in delay
• Zurich’s “insert and return”
34
Passive Priority
• Treatments that favor buses and that don’t
rely on bus detections
• Favorable splits for bus phase
• Favorable offsets (progression) for bus
– Hard to do over more than a few intersections
due to uncertain dwell time
• Short cycles or double realizations (short
red is the key)
35
Ruggles Bus Terminal Study
Study
with Burak Cesme, PhD student at
Northeastern Univ.
Using VISSIM simulation and VAP signal
controll programmiing
36
Figure by MIT OpenCourseWare.
Bus Del
B
D lays with
ith
Incremental Priority
Treatments, by Route
Figure by MIT OpenCourseWare.
38
Passive priority: more time to bus left turn. Note: 5 s increase in split consumes only 2.5
2 5 s! Max Green = 21 seconds
00.90
90
0.80
0.70
0.60
0.50
0.40
0.30
0 20
0.20
0.10
0.00
Proportion
Pr
roportion
Pr
roportion
Max Green = 16 seconds
8
9
10
11
12
13
Green Time ( s )
14
15
16
00.90
90
0.80
0.70
0.60
0.50
0.40
0.30
0 20
0.20
0.10
0.00
8
9 10 11 12 13 14 15 16 17 18 19 20 21
Green Tim e ( s )
Avg Green (EBL) = 15.3s
15 3s
p (max-out) = 84.6%
Avg Green (EBL) = 17.8s
17 8s
p (max-out) = 51.2%
Avg bus delay = 98 s
Avg bus delay = 67 s
39
39
Intelligent Green Extension: 10 s extension for “cost”
cost of 0.5
0 5 s
s
With Green Extension
0.60
060
0.60
0.50
0.50
0.40
0.40
Prroportion
Prroportion
No Priority
0.30
0.20
010
0.10
Extended Green
0.30
0.20
0.10
0.00
0.00
8 9 10 11 12 13 14 15 16 17 18 19 20 21
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Green Time ( s )
Green Time ( s )
p (max-out) = 0.512
Avg Green (EBL) = 18.1s
p (max
(max-out)
out) = 0.247
0 247
p (extended) = 0.213
Avg
gGreen (WBT))=
30.3s
AvggGreen (WBT)) =
29.8s
Avg Green (EBL) = 17.8s
40
Inserting 10 s phase: only consumes 2.5 s
Primary Phase , when Phase Insertion is
Programmed
Green Extension Only
0.30
0.30
0.25
0.25
Extended Green
0.15
0.20
Proportion
n
Proportion
n
0.20
0.10
0.05
0 00
0.00
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
Extended Green
0.15
0.10
0.05
0 00
0.00
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31
GreenT ime ( s )
Green Time ( s )
Avg Green (EBL) = 18.1s
Bus delay = 55 s
Avg Green (WBT) = 29.8s
p (insertion)
(i
ti ) = 0.386
0 386
Avg Green (primary) = 14.7s
Avg Green (insertion) = 4.4s
A G
Avg
Green (t
(total)
t l) = 19.1s
19 1
Bus delay = 33 s
Avg Green (WBT) = 27.3s
41
Conditional Priority
Priority to Late Buses
•
•
•
•
•
Less interference with traffic (Eindhoven)
Push pull means of operational control (Einhoven)
Push-pull
What is “Late:” 15 s or 3 minutes?
Demands fine-tuned schedule
y for short-headwayyservice
Headwayy-based priority
42
Priority Queue Management
• Detectors & logic for queue management
– Stopped cars, not moving cars, hinder buses
(Zuriich)
(Z
h)
(Eindhoven)
43
Z i h’ Custom
Zurich’s
C t
Programming
P i
• 5 full-time programmers work on signal
control programs
• Logic runs in central computers; field
controllers
ll merely
l implement
i l
& communicate
i
• Exp
perience has taugght them:
– Delay tram green until trams start to slow down
–Evaporated
Evaporated traffic
– Early red for the safety of last-moment crossing
pedds
44
Prediictive
i Priioriity
Remote, upstream detection: simulated on Huntington Ave, used in Salt Lake City
• D
Detect
t tor 1 used
d to
t predi
dictt bus
b arriivall att 4 (~2
( 2 miinuttes
advance)
• Adjust cycle lengths so that bus will arrive on green
• Last-minute priority as backup
• Adaptive
Adapti e (learning) algorithm for predicting bus
b s arrival
arri al
45
Dynamic Coordination
Coordination
((Zurich))
–S
Small
ll zones (1-3
(1 3 intersections)
i t
ti )
– No fixed clock
– Shape green waves through the zone around
bus
– Zone boundaries are segments that offer storage
bu e
buffer
46
Self-Organizing Coordination
Simulated for San Juan, Puerto Rico
• Each signal’s start of green becomes a request to
downstream signal
– Peer-to-peer communication between signals
– upstream
t
signal’s
i l’ requestt has
h lower
l
priority
i it that
th t bus
b
request
• Result
lt: sponttaneous green wave
• Inherently interruptible
47
M l i Levell Pri
Multi-L
P iority
i
(South Tangent = Haarlem – Airport – Amsterdam South)
• Bus is early: green extension only
• 0 to3 min late: “normal” early green
• More than 3 min late: aggressive early
green (skip competing phases)
48
Bus Delays
with
Incremental
Priority
Treatments, Treatments
by Route
Figure by MIT OpenCourseWare.
49
Six Keys to Perfformance
Si
1. Aim for near-zero delay
2. Multiple intelligent and aggressive tactics, with
with
compensation
3. Coordinate with scheduling (cond
(cond’ll priority)
priority)
4. Alternatives to rigid coordination
5. Advanced prediction with gradual cycle
adjustments
6. Custom programming, performance
measurement, & continual improvement
50
MIT OpenCourseWare
http://ocw.mit.edu
1.258J / 11.541J / ESD.226J Public Transportation Systems
Spring 2010
For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.