Transit Signal Priority Experiment in a Connected Vehicle Technology Environment

Abstract

Connected vehicle (CV) technology allows developing next-generation transit signal priority (TSP) that is based on cooperation between signals (change signal timing) and buses (alter speed). In this study, a CV-based TSP that was previously developed and evaluated through a computer simulation was tested in the field. The field experiment at the Smart Road testbed at the Virginia Tech Transportation Institute (Blacksburg, Virginia) validated the feasibility of CV-based TSP and evaluated its performance. The proposed TSP algorithm properly delivered TSP green times to buses with a 100% success rate. In addition, it reduced delays—between 32 and 75%—for a bus traveling approximately 72.4 km=h (45 mi=h) and a traffic signal with a 90-s cycle length with 30-s green time. Moreover, the field experiment showed that regular and differential global positioning system (GPS) devices demonstrate no statistically significant difference in performance. This finding could facilitate large-scale implementation of the TSP logic based on connected vehicle technology (TSPCV) because the regular GPS devices are readily available and much cheaper than differential GPS devices.

Full text

Case Study
Transit Signal Priority Experiment in a Connected Vehicle
Technology Environment
Young-Jae Lee, Ph.D.1; Seyedehsan Dadvar, Ph.D., S.M.ASCE2; Jia Hu, Ph.D., A.M.ASCE3;
and Byungkyu Brian Park, Ph.D., M.ASCE4
Abstract: Connected vehicle (CV) technology allows developing next-generation transit signal priority (TSP) that is based on cooperation
between signals (change signal timing) and buses (alter speed). In this study, a CV-based TSP that was previously developed and evaluated
through a computer simulation was tested in the field. The field experiment at the Smart Road testbed at the Virginia Tech Transportation Institute
(Blacksburg, Virginia) validated the feasibility of CV-based TSP and evaluated its performance. The proposed TSP algorithm properly delivered
TSP green times to buses with a 100% success rate. In addition, it reduced delays—between 32 and 75%—for a bus traveling approximately
72.4km=h(45 mi=h) and a traffic signal with a 90-s cycle length with 30-s green time. Moreover, the field experiment showed that regular and
differential global positioning system (GPS) devices demonstrate no statistically significant difference in performance. This finding could facili-
tate large-scale implementation of the TSP logic based on connected vehicle technology (TSPCV) because the regular GPS devices are readily
available and much cheaper than differential GPS devices. DOI: 10.1061/JTEPBS.0000062.© 2017 American Society of Civil Engineers.
Author keywords: Transit signal priority; Connected vehicle; Field experiment; Global positioning system (GPS); TSP logic based on
connected vehicle technology (TSPCV).
Introduction
For years, transit signal priority (TSP) has been proposed and stud-
ied as an efficient way of improving transit operations. It offers
preferences to transit vehicles at traffic signalized intersections and
has been proven valuable in reducing transit travel time and im-
proving schedule adherence and customer ride quality (Pratt et al.
2000). Furthermore, it has been shown that TSP has the ability to
mitigate the negative effects stemming from outdated timing plans
(Muthuswamy et al. 2007). The technology has been applied in
many cities in Europe, Asia, and North America. In the United
States, Seattle, Portland, Oregon, Los Angeles, Chicago, and many
other large cities have implemented the TSP system (Liao and
Davis 2007).
Past studies in the United States showed that the benefits of TSP
in terms of bus travel time savings vary significantly, as noted in
Table 1. Among the reviewed quantitative TSP studies, the travel
time savings ranged between 0.9 and 71%. The past studies can be
categorized into two main categories: simulation-based evaluations
and field tests. Due to difficulties associated with field tests, the
majority of past studies were simulation-based. Some studeis were
based on conventional TSP (CTSP) like Rakha and Ahn (2006),
which evaluated TSP deployed in Arlington, Virginia. The saving
on bus travel time was 0.9% but resulted in 1% increase on total
delay. A closer investigation of that particular TSP system indicates
at least two possible reasons why the benefit was marginal. First,
TSP logic was too simple (i.e., only a green extension of 5 s).
Second, the progression between adjacent intersections was not co-
ordinated. In recent years, there have been some notable studies
relying on both simulation and field tests like the Multi-Modal
Intelligent Traffic Signal System (MMITSS) project [Ahn et al.
2015;University of Arizona, University of California PATH
Program, Savari Networks, Inc., Econolite 2016] that aimed at im-
plementing a multimodal intelligent traffic signal system within a
connected vehicle environment. This MMITSS project investigated
TSP on a fairly high-level basis and was a valuable guideline for
TSP research; however, the project did not provide detailed TSP
algorithm innovation.
In addition to the previously mentioned shortcomings, several
other challenges prevent the conventional TSP from being widely
deployed. One potential challenge of the conventional TSP is to
predict the bus arrival time (Rakha and Ahn 2006). Because of the
uncertainty of the bus’s arrival time, the TSP procedure usually
moves a large portion of the green time from the side streets to the
street where a bus is expected to arrive. In some worst cases, a bus
would arrive in the next cycle without taking advantage of any of
the extended green time, while the vehicles on the side street keep
waiting and accumulating delay time. Obviously, this causes a sig-
nificant adverse effect on traffic conditions.
To properly address these challenges, a next-generation TSP
logic based on connected vehicle technology (TSPCV) was pro-
posed in Hu et al. (2014,2015). This new TSP takes advantage of
the resources provided by connected vehicle (CV) technology, in-
cluding two-way communications between the bus and the traffic
signal controller, accurate bus location detection, and prediction and
the number of passengers. The key feature of the proposed TSP
1Associate Professor, Dept. of Transportation and Urban Infrastructure
Studies, Morgan State Univ., 1700 E. Cold Spring Ln., Baltimore,
MD 21251. E-mail: YoungJae.Lee@morgan.edu
2Graduate Research Assistant, Dept. of Transportation and Urban Infra-
structure Studies, Morgan State Univ., 1700 E. Cold Spring Ln., Baltimore,
MD 21251. E-mail: Seyedehsan.Dadvar@morgan.edu
3Research Associate, Turner Fairbank Highway Research Center,
Federal Highway Administration, 6300 Georgetown Pike, McLean, VA
22101 (corresponding author). ORCID: http://www.orcid.org/0000-0002
-0900-7992. E-mail: jh8dn@virginia.edu
4Associate Professor, Dept. of Civil and Environmental Engineering,
Univ. of Virginia, P.O. Box 400742, Charlottesville, VA 22904-4742.
E-mail: bpark@virginia.edu
Note. This manuscript was submitted on April 27, 2016; approved on
February 1, 2017; published online on May 3, 2017. Discussion period
open until October 3, 2017; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Transportation En-
gineering, Part A: Systems, © ASCE, ISSN 2473-2907.
© ASCE 05017005-1 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Table 1. Summary of Past TSP Studies
Location
Number of
buses Tool Arrival time forecast TSP type Measurements Result Reference
Ann Arbor,
Michigan
1TRAF-NETSIM Historical data and
sensitivity analysis
Conventional Bus travel time 13.5% reduction Khasnabis et al.
(1996)
Ann Arbor,
Michigan
1NetSim Check in, check out Conventional Delay Bus Little benefit Al-Sahili and
Taylor (1996)Automobile Increase
Austin, Texas 1 TRAF-NETSIM Not clear Conventional Off-peak travel time 11% improved Garrow and
Machemehl (1997)Peak time travel time TSP was not effective
Hypothetical
network with three
intersections
1TexSIM Not clear TSP with AVL Stop delay Vehicles in bus’s direction −6to 10% Balke et al. (2000)
Vehicles on cross street þ2to 26%
Portland, Oregon Six bus
routes
Network data
(before and after
TSP)
N/A Conventional Run time savings 45.8% improved Kimpel et al.
(2005)On-time performance 41.7% improved
Mean headway 37.5% improved
Hypothetical
intersection
1NetSIM N/A Adaptive TSP Total delay −3to 71% Lee et al. (2005)
Arlington, Virginia 1 INTERGRATION +
field test
Average from historical
global positioning
system data
TSP with AVL Reliability 3.20% Rakha and Ahn
(2006)Bus travel time −0.90%
Total delay Per vehicle 1%
Per person 0.60%
Newark, New Jersey 1 WatSIM Historical data, no AVL Conventional Travel time Bus −10 to 20% Muthuswamy et al.
(2007)Automobile −5to 10%
(main street)
Minneapolis 1 Aimsun Average from historical
global positioning
system data
TSP with AVL Bus travel time Morning peak −12 to 15% Liao and Davis
(2007)Evening peak þ4to 11%
Snohomish County,
Washington
1 Field-observed
data and VisSim
TSP reader logs Conventional Transit time match 6–16.3% improved Wang et al. (2008)
Transit travel time 5% improved
Traffic queue length Insignificant
Signal cycle failures Insignificant
Average person delay Decreased
Vehicle delays and stops Insignificant
Fortaleza, Brazil Single and
multiple
SCOOT On-board surveys based
on AVL data
TSP with AVL Travel time Bus 10–20% reduction Oliveira-Neto et al.
(2009)Automobile
Vancouver, Canada 1 VisSim Linear model fit by past
data
TSP with AVL Bus travel time −33% Ekeila et al. (2009)
Cross-street delay Negligible
Beiyuan Road,
Ji’nan City, China
1 Field test Not clear Coordinated and
conditional bus
priority (CCBP)
Bus delay 34.7% reduction Ma et al. (2010)
Total average delay of motor vehicles 8.9% increase
Central Avenue,
Minneapolis
1 Field test Real-time bus arrival
time prediction based on
historical data with AVL
TSP with AVL Travel time reduction 3–6% Liao and Davis
(2011)
Speedway
Boulevard, Tucson,
Arizona
Multiple
buses
VisSim Real-time bus arrival
time computation based
on global positioning
system and CV data
TSPCV Average bus delay in congested conditions Approximately 50%
reduction
He et al. (2011)
Anthem, Arizona Multiple
modes
VisSim and
field test
Real-time arrival time
computation based on
global positioning
system and CV data
TSPCV Travel time Improved Ding et al. (2013)
© ASCE 05017005-2 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Table 1. (Continued.)
Location
Number of
buses Tool Arrival time forecast TSP type Measurements Result Reference
Hypothetical
intersection
1 Field test and
unknown simulation
tool
N/A TSPCV (CVIS) DSRC performance Meets the basic
requirements
Wang et al. (2013)
DGPS positioning performance 98% accuracy
(0.1m=s)
Average travel time, stopping time, and
queue length
All improved in
nonsaturated
conditions
El Camino Real,
northern California
1Paramics Average from historical
data and sensitivity
analysis
TSP with AVL Improvements in CO2emissions All vehicles: ∼1% Yelchuru et al.
(2014)Bus: ∼1.5%
Improvements in delay All vehicles: ∼1.5%
Bus: ∼3%
Speedway
Boulevard, Tucson,
Arizona
Multiple
modes
VisSim Predefined rates and
times
TSPCV Bus delay 24.9% improved He et al. (2014)
Pedestrian delay 14% improved
Car delay No change
Taicang City, China 1 Field test Real-time bus arrival
time computation based
on global positioning
system and CV data
TSPCV Travel time 33–40% improved Wang et al. (2014)
Number of stops Approximately 80%
improved
Arizona Connected
Vehicle Test Bed
1 Field test and VisSim Real-time arrival time
computation based on
global positioning
system and CV data
Combination of TSP
and freight signal
priority MMITSS
Transit travel time 6.1–8.2% improved Ahn et al. (2015)
Transit delay Up to 10.5% reduced
Arizona Connected
Vehicle Test Bed,
California
Connected Vehicle
Test Bed
1 Field tests and
VisSim
Real-time arrival time
computation based on
global positioning
system and CV data
TSPCV (MMITSS)
and freight signal
priority
Functionality Passed (field tests) University of
Arizona,
University of
California PATH
Program, Savari
Networks, Inc.,
Econolite (2016)
Number of stops at red 7.3% improved
Cumulative time of delay 14.1% improved
Trip travel time 7.4% improved
Cumulative intersection delay 12.5% improved
Note: AVL = automatic vehicle location; CVIS = cooperative vehicle-infrastructure systems; DGPS = differential global positioning system; DSRC = dedicated short range communication.
© ASCE 05017005-3 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

logic is the green time reallocation, which moves green time instead
of adding extra green time. The TSP was also designed to be condi-
tional. That is, the delay per person served as a criterion determin-
ing whether TSP is granted. Based on the simulation results, the
proposed TSP accommodates a higher percentage of transit buses
than conventional TSP. Its performance was compared against the
CTSP and no-TSP (NTSP) cases under various congestion levels.
The results showed that the TSPCV greatly reduced the bus delays
at traffic signalized intersections without causing statistically sig-
nificant negative effects on side streets.
Objectives
To the study team’s best knowledge, a minority of reviewed studies
(Table 1) investigated the performance benefits of TSP based on the
field test. Among the field tests, only a limited number had the ex-
periment under a CV environment (Ma et al. 2010;Ding et al. 2013;
Wang et al. 2014;Ahn et al. 2015;University of Arizona, Univer-
sity of California PATH Program, Savari Networks, Inc., Econolite
2016) in China (Ji’nan City and Taicang City) and the United States
(CV test beds in Arizona and California). The U.S.-based studies
have been done almost in parallel with this study, so this study is
one of the few TSP studies that validate the TSP performance by
field experiment, and one of the first TSP experiments with an in-
telligent TSP environment.
The TSP experiment in this study aims at validating the pro-
posed TSPCV algorithm in Hu et al. (2014,2015) within a CV envi-
ronment and estimating selected performance measures. It can
assure the implementation of the proposed TSPCV at the Smart
Road of the Virginia Tech Transportation Institute (Blacksburg,
Virginia), confirm software and hardware compatibility, illuminate
TSPCV performance, and reveal global positioning system (GPS)
requirements (regular and differential). This study will be the first
TSPCV field experiment at the Smart Road of the Virginia Tech
Transportation Institute.
Methodology
A bus can arrive at a green, yellow, or red traffic signal depending
on traffic signal phasing and bus speed, which is mainly subject to
•Roadway geometry;
•Roadway speed limit; and
•Speed of other vehicles in front of the bus on the road upstream
of the intersection.
Fig. 1shows possible arrival times at 0.8 km (0.5 mi) [i.e., the
distance of roadside equipment (RSE) to the intersection at the
Smart Road at the Virginia Tech Transportation Institute] from
the traffic signalized intersection with the assumption that the bus
maintains its speed at 0.8 km (0.5 mi) until it reaches the inter-
section. In those cases, the bus can pass the intersection during
the original green time or even yellow time, and then there is no
place for TSP. However, when the bus arrives during the red time
of the original signal phasing, then TSP can be beneficial and
provide the bus a short green time to pass the intersection without
waiting for the next green time of the original signal phasing
(Fig. 1).
Based on the TSPCV logic architecture, which has been de-
scribed in Hu et al. (2014,2015), key components of TSPCV for
this experiment have been defined as follows (Fig. 2):
1. Bus detection component:
a. An approaching vehicle [an Infiniti FX35 (2005) (Nissan
Motor, Japan) enabled with CV features] that resembles a
bus passes the activation point [0.8 km (0.5 mi) to the
intersection, which is the distance of RSE (Savari Street-
Wave, Savari) to the intersection] and activates the TSP
algorithm.
(1) Bus speed and location are measured with two GPS
devices: regular GPS [GPS from NextGen Head Unit,
Virginia Tech Transportation Institute (VTTI), Blacks-
burg, Virginia] and differential GPS [Novatel Flexpak6
(NovAtel, Canada) located in vehicle trunk].
(2) The communication between the bus and RSE is
provided via on-board equipment (OBE) [Savari OBE
S100 (Savari, United States) located in trunk].
(3) Parametric data are collected in the NextGen data acqui-
sition system [(DAS), Virginia Tech Transportation Insti-
tute (VTTI), Blacksburg, Virginia] such as GPS position,
GPS speed, network speed, turn signal, brake, accelerator
position, and rotations per minute (rpm).
(4) The algorithm is running on a laptop [Dell Latitude
E6430s (Dell, Round Rock, Texas), Core i5 vPro], which
is connected to the vehicle’s GPS device and communi-
cates with RSE and the traffic signal controller [custom
proprietary interface with a D4 traffic signal controller
(Fourth Dimension Traffic Company)].
Fig. 1. Possible arrival times at 0.8 km (0.5 mi) to a signalized intersection and TSP green time allocation
© ASCE 05017005-4 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Distance (0.8 km + d) 0.8 km (0.5 miles) 0
d
RSE
TSP
GPS and Other
OBE Devices
RSE
TSP Algorithm
(Laptop)
Fig. 2. TSPCV experiment structure and experiment site at Virginia Tech Transportation Institute (Smart Road) (map data ©2015 Google
and ESRI).
© ASCE 05017005-5 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

b. After TSP activation, the algorithm checks the state of the bus
and expected arrival time at intersection.
c. If the bus cannot reach the original green time, then the TSP
algorithm will modify the signal timing and the system pro-
ceeds to the next step. Otherwise, the TSP process is termi-
nated and signal timing will not be modified.
2. TSP timing plan and bus speed calculation component:
a. The proposed algorithm generates a timing plan and calcu-
lates the corresponding recommended bus speed.
b. Advisory bus speed is calculated by solving a binary mixed-
integer linear program (BMILP) problem. Detailed informa-
tion about formulation of the BMILP problem is provided in
Hu et al. (2014,2015).
3. Logic assessment and implementation component:
a. The TSP timing plan is compared against the normal signal
time (the signal schedule with lower per person delay is im-
plemented) and the recommended bus speed will be trans-
mitted to the bus. It can be either announced via speaker or
read by a study team member through the display screen
[high-definition multimedia interface Feelworld 5-in. thin-
film-transistor liquid crystal display monitor (Feelworld,
Germany)]. Instructions are given to a bus driver about the
recommended speed.
b. A buffer green time is possibly given to a bus in case a bus is
not expected to make it through the intersection. The TSP
green time would be extended up to 5 s to accommodate the
random delay. This buffer green time is referred to as green
extension in the following sections. All required computation
is running on the laptop inside the vehicle.
The ddistance in Fig. 2is the required distance at which the
vehicle can achieve the target speed at the 0.8-km (0.5-mi) point.
The distance was identified via trial and error based on the experi-
ment roadway geometry and selected target speed for the bus.
Fig. 3demonstrates a data-flow diagram of the TSPCV experi-
ment between system components including vehicle and GPS de-
vices, RSE at 0.8 km (0.5 mi) to the intersection, RSE and traffic
signal controller located at intersection, and the central server. The
following are the main flows of data and communication:
•Original signal phasing from the traffic signal controller to
TSPCV on the laptop in the vehicle (solid arrows);
•Data from either regular GPS or differential GPS to TSPCV on
the laptop in the vehicle (dashed arrows);
•TSPCV signal phasing from TSPCV on the laptop in the vehicle
to the traffic signal controller (dash-dot arrows); and
•TSPCV logic (dotted arrows).
The signal phasing is coded as follows:
1. Red;
2. Yellow; and
3. Green.
The TSPCV algorithm receives vehicle location and speed from
the GPS devices (either regular or differential) as well as current
(original) signal phasing from the traffic signal controller, and,
based on its logic of whether the vehicle is behind schedule, sends
back appropriate TSPCV signal phasing (translated into Codes 1, 2,
and 3) to the intersection signal controller. If there is no need
for TSPCV signal phasing implementation, then current (original)
signal phasing continues.
Experiment
Experiment Site
Fig. 2shows the experiment site. The line on the map is the Smart
Road at the Virginia Tech Transportation Institute. The study team
started each trial of the experiment at the start point, which was
approximately 1.05 km (0.65 mi) to the intersection at the Smart
Road. The study team could identify the value of d(Fig. 2) as ap-
proximately 0.25 km (0.15 mi). Starting each trial at this point
ensured reaching the target speed [72.4km=h(45 mi=h), which
will be explained in the next section] at 0.8 km (0.5 mi) from the
intersection (TSPCV activation point) in each scenario.
Experimental Scenarios
The following scenarios were of interest to the study team to test.
Signal Phasing
Transit signal priority is most effective in an urban or suburban area
on a major arterial where traffic is heavy. The study team decided
to include a short green time for the experiment’s signalized inter-
section because it could induce more situations to validate the pro-
posed TSPCV by having a long red time. Therefore, the experiment
assumed the following:
•Cycle length of 90 s including green time of 30 s, yellow time of
3 s, all red time of 2 s, and red time of 55 s; and
•Traffic signal cycle length and the sequence of signal phases
were fixed.
Arrival Types
The reliability of the TSPCV mechanism is investigated here. It
demonstrates whether TSPCV could perform properly under differ-
ent activation scenarios. One major difficulty for TSP is that the
arriving bus could conflict with the minimum green time require-
ment of the other three approaches of the intersection. As was
stated, there are different possibilities for the bus arrival times dur-
ing the red time of original signal phasing in Fig. 1. Whether or not
the mechanism could lead the vehicle to avoid the minimum green
time window and successfully provide the TSP green phase is the
question to address here. To measure the performance of the pro-
posed TSPCV at different arrival times of red time, the study team
categorized them into the following five groups; trials that started at
either 40 or 50 or 60 or 70 or 80 s of original signal cycle.
Table 2summarizes the experiment scenarios and number of
trials for each scenario that the study team tested at the Smart
Road of the Virginia Tech Transportation Institute. The experiment
included 36 trials. As previously mentioned, providing different
arrivals during red time was one of the key elements for the experi-
ment. So after a few trials and errors, it was found that starting at a
point approximately 1.05 km (0.65 mi) from the intersection (Fig. 2)
could make the driver reach approximately 72.4km=h(45 mi=h) at
the 0.8-km (0.5-mi) point from the intersection, and the bus could
reach the intersection in approximately 55 s by maintaining approx-
imately 72.4km=h(45 mi=h). Then, to assure different arrivals
during the red time arrival times, the bus started the traffic signal
at different points of the original signal cycle at the same time as the
vehicle was starting to move at the point approximately 1.05 km
(0.65 mi) from the intersection (i.e., Start Point in Fig. 2). Thus,
“Cycle length start time”in Table 2refers to the point of original
signal cycle that was started by study team just as the vehicle had
started to move at 1.05 km (0.65 mi) from the intersection.
GPS Devices
The following were the GPS devices used:
•Regular GPS (GPS from NextGen Head Unit).
•Differential GPS (Novatel Flexpak6).
Differential GPS devices are more accurate than regular GPS
devices but cost much more. The purpose was to evaluate the ben-
efit of extra cost associated with the differential GPS.
© ASCE 05017005-6 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Speed Limit
Again, because the TSPCV is being tested under a major arterial
context, the most commonly posted speed limit is adopted,
72.4km=h(45 mi=h).
Equipment Examination
Prior to the experiment, the study team visually confirmed the com-
patibility of the algorithm and the equipment (OBE and RSE) at the
Virginia Tech Transportation Institute.
Experiment Data Collection
From the beginning of each trial to the end (a few meters after pass-
ing the intersection), all experiment data were recorded for each
second of each trial. The critical measurements collected were as
follows:
•Time;
•Coordinated universal time (UTC);
•Original timing plan;
•TSP timing plan [activated after the bus passed 0.8 km
(0.5-mi) point];
•Bus speed;
•Bus location;
•Distance to intersection; and
•Traffic light status at intersection [prior to 0.8 km (0.5 mi) loca-
tion was based on original timing plan and then TSP timing plan
after the vehicle passes the 0.8 km (0.5 mi) location].
After the experiment, the team compiled all recorded data and
performed statistical analyses as presented in the following sections.
Analysis
The evaluation of the TSPCV experiment was based on the success
rate, relationship between delay reduction and original red light
arrival time, and GPS type effect. The following presents the for-
mulation related to delay computation.
In order to compute delay, predicted arrival time without
TSPCV needs to be first calculated. Because at each trial—at a
starting point approximately 1.05 km (0.65 mi) from the intersec-
tion (Fig. 2), at the same time as bus movement—the study team
started the operation of the traffic signal at a predefined point of
the signal cycle (Table 2); later, when all experiment data were
Fig. 3. TSPCV experiment data flow diagram
Table 2. TSP Experiment Scenarios
Scenarios Regular GPS Differential GPS
Speed limit [km/h (mi/h)] 72.4(45)72.4(45)
Cycle length
start time
40 s×10 trials 40 s×3trials
50 s×3trials 50 s×2trials
60 s×3trials 60 s×3trials
70 s×3trials 70 s×3trials
80 s×3trials 80 s×3trials
Subtotal 22 trials 14 trials
Total 36 trials
© ASCE 05017005-7 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

recorded, the study team could predict the arrival time (from the
aspect of corresponding original cycle length) of the bus at the in-
tersection without TSP activation at 0.8 km (0.5 mi) from the
intersection using the following formula, and called this value the
predicted relative cycle length arrival time without TSP at 0.8 km
(0.5 mi) (which is shown in tables in subsequent sections):
PRCAT ¼D
VþCST−ðn×CÞð1Þ
where PRCAT = predicted relative cycle length arrival time without
TSP at 0.8 km (0.5 mi); V= bus speed (m=s); D= distance to in-
tersection (m) (0.8km ¼0.5mi); CST = cycle length start time (s);
n= number of passed signal cycles (if bus arrives at the same signal
cycle then n¼0, the next signal cycle then n¼1, and so on); and
C= original cycle length without TSP (90 s).
The value of PRCAT falls at any number between 0 and 90; from
0 to 55, the traffic light would be red, from 55 to 85, it would be
green, from 85 to 88, it would be yellow, and from 88 to 90, it
would be all red. But as mentioned, after a few trials and errors, the
study team could identify the range of the cycle length start time in
which the bus would arrive at different points of the original red
time (i.e., cycle length start time = either 40, 50, 60, 70, or 80 s).
Other delay-related measurements of effectiveness are presented
in the following:
Delayw=oTSP ¼POT −D
V
×3,600 ð2Þ
DelayTSP ¼AOT −D
V
×3,600 ð3Þ
where Delayw=oTSP = delay without TSP (s); DelayTSP = delay with
TSP (s); POT = predicted overall time to pass the intersection with-
out TSP (s); AOT = actual overall time to pass the intersection with
TSP (s); V= bus speed (m=s); and D= distance to intersection (m)
(0.8km ¼0.5mi)
Reduced DelayðsÞ¼Delayw=oTSP −DelayTSP ð4Þ
Reduced Delayð%Þ¼ðDelayw=oTSP −DelayTSPÞ=ðDelayw=o TSPÞ
ð5Þ
Evaluation
Success Rate
Table 3summarizes the success rate for the proposed TSPCV
algorithm for different tested scenarios. Generally, the proposed
method could provide the green time for the bus 100% of the
time. Approximately 50% of the time (for both regular and
differential GPS devices), the bus operated without green
extension.
Relationship between Delay Reduction and Original
Red Light Arrival Time
Table 4shows the amount of delay time reduced by TSPCV in com-
parison with no TSPCV for average performance of each set of trials
of each scenario. Fig. 4summarizes the values of reduced delays in
seconds and percentages. As expected, TSP could save more time
when the bus arrived at the beginning of the red light signal phasing
because it could avoid longer red light signal timing for the bus.
However, when the bus arrived later (midred or late red), the overall
saved time decreased accordingly. Overall, an average of 57% delay
is reduced with TSPCV. It confirms the simulation study conducted
by Hu et al. (2014,2015).
Table 3. Success Rates for Different Relative Cycle Length Arrival Times
Scenario
Cycle length
start time
Number
of tests
Number of
TSP green
provided
Success
rate (%)
Number of TSP
green phase with
shorter delay
TSP green phase
with shorter
delay (%)
Number of TSP green
phase with shorter
delay without
green extension
TSP green phase
with shorter delay
without green
extension (%)
Regular GPS 40 10 10 100 10 100 4 40
50 3 3 100 3 100 2 67
60 3 3 100 3 100 1 33
70 3 3 100 3 100 1 33
80 3 3 100 3 100 2 67
All 22 22 100 22 100 10 45
Differential GPS 40 3 3 100 3 100 2 67
50 2 2 100 2 100 2 100
60 3 3 100 3 100 0 0
70 3 3 100 3 100 3 100
80 3 3 100 3 100 0 0
All 14 14 100 14 100 7 50
Table 4. Reduced Delays for Different Red Light Arrival Times
Scenario
Cycle
length
start time
(s) Trial
Predicted
relative cycle
length arrival
time without
TSP at 0.8 km
(0.5 mi)
Reduced
delay
(s)
Reduced
delay
(%)
Regular
GPS
40 Average 5.4 38.7 60
50 Average 16.9 38.3 70
60 Average 25.4 29.0 64
70 Average 35.3 17.7 50
80 Average 44.7 9.7 39
Differential
GPS
40 Average 1.9 41.3 64
50 Average 11.9 41.0 75
60 Average 24.3 27.0 60
70 Average 30.3 21.7 62
80 Average 45.3 8.0 32
© ASCE 05017005-8 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

GPS Type Effect
Regular GPS
In this scenario, the vehicle enabled with a regular GPS device was
driving at approximately 72.4km=h(45 mi=h) at 0.8 km (0.5 mi)
from the intersection and then, depending on predicted arrival time
at the intersection, the driver was advised by the algorithm to main-
tain the speed that could lead the vehicle to pass the intersection
during a TSP-provided green phase. Table 5summarizes this sce-
nario and its trials. The proposed TSP succeeded in giving the
vehicle the green light in all trials. On average, the amount of delay
reduced by the TSP algorithm varied from 9.7 s or 39% (for arriv-
ing almost at the end of the red light cycle of the intersection) to
38.7 s or 70% (for arriving almost at the early red light cycle of the
intersection).
Because it was the first set of trials, the team tried the first trial
(starting at 40 s of original intersection signal phasing at the start
point) more than all other trials to get familiar with the performance
of the TSP algorithm. However, the algorithm worked fine for all
scenarios.
0.0
10.0
20.0
30.0
40.0
50.0
0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 45.0 50.0
PSThtiwyaleDdecudeR
(s)
Predicted Relative Original Red Light Arrival Time w/o TSP (s)
72.4 km/h (45 mi/h) [Regular GPS] 72.4 km/h (45 mi/h) [Differential GPS]
0%
20%
40%
60%
80%
0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 40.0 45.0 50.0
)%(PSThtiwyaleDdecudeR
Predicted Relative Original Red Light Arrival Time w/o TSP (s)
72.4 km/h (45 mi/h) [Regular GPS] 72.4 km/h (45 mi/h) [Differential GPS]
(a)
(b)
Fig. 4. Reduced delays for different original red light arrival time: (a) reduced delays in seconds; (b) reduced delays in percentage
Table 5. Regular GPS Scenario
Cycle
length
start
time
(s) Trial
Speed at
0.8 km
(0.5 mi)
[km=h (mi=h)] PRCAT
Predicted
original traffic
light status
without TSP
Predicted
overall
time to
pass the
intersection
without TSP
Actual
traffic
light
status
with
TSP
Actual
overall
time to
pass the
intersection
with TSP
Green
extension
(s)
Delay
without
TSP
(s)
Delay with
TSP
(s)
Reduced
delay
(s)
Reduced
delay
(%)
40 1 69.7 (43.3) 7.3 Red 105.0 Green 63.0 0 65.0 23.0 42.0 65
2 73.2 (45.5) 4.6 Red 105.0 Green 61.0 0 65.0 21.0 44.0 68
3 72.3 (44.9) 5.6 Red 105.0 Green 65.0 2 65.0 25.0 40.0 62
4 71.5 (44.4) 3.3 Red 105.0 Green 63.0 0 65.0 23.0 42.0 65
5 74.2 (46.1) 90.0 Red 105.0 Green 63.0 0 65.0 23.0 42.0 65
6 69 (42.9) 7.0 Red 105.0 Yellow 80.0 5 65.0 40.0 25.0 38
7 69.7 (43.3) 5.9 Red 105.0 Green 66.0 3 65.0 26.0 39.0 60
8 67.1 (41.7) 8.3 Red 105.0 Yellow 70.0 5 65.0 30.0 35.0 54
9 66.6 (41.4) 8.0 Red 105.0 Green 67.0 4 65.0 27.0 38.0 58
10 67.6 (42) 3.9 Red 105.0 Green 65.0 2 65.0 25.0 40.0 62
50 1 65 (40.4) 19.4 Red 95.0 Green 58.0 2 55.0 18.0 37.0 67
2 69.5 (43.2) 15.8 Red 95.0 Green 56.0 0 55.0 16.0 39.0 71
3 70.5 (43.8) 15.4 Red 95.0 Green 56.0 0 55.0 16.0 39.0 71
60 1 69.7 (43.3) 25.9 Red 85.0 Green 56.0 0 45.0 16.0 29.0 64
2 71.5 (44.4) 24.2 Red 85.0 Green 56.0 1 45.0 16.0 29.0 64
3 70.5 (43.8) 26.1 Red 85.0 Green 56.0 0 45.0 16.0 29.0 64
70 1 72.4 (45) 35.5 Red 75.0 Green 58.0 1 35.0 18.0 17.0 49
2 69.4 (43.1) 36.5 Red 75.0 Green 59.0 3 35.0 19.0 16.0 46
3 73.5 (45.7) 34.0 Red 75.0 Green 55.0 0 35.0 15.0 20.0 57
80 1 71.5 (44.4) 44.0 Red 65.0 Green 54.0 0 25.0 14.0 11.0 44
2 69 (42.9) 47.0 Red 65.0 Green 56.0 0 25.0 16.0 9.0 36
3 73.2 (45.5) 43.2 Red 65.0 Green 56.0 1 25.0 16.0 9.0 36
© ASCE 05017005-9 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Differential GPS
In this scenario, the vehicle enabled with the differential GPS de-
vice was driving at approximately 72.4km=h(45 mi=h) at 0.8 km
(0.5 mi) from the intersection and then, depending on predicted
arrival time at the intersection, the driver was advised by the
algorithm to maintain the speed that could lead the vehicle to pass
the intersection at a TSP-provided green light. Table 6summarizes
this scenario and its trials. The proposed TSP succeeded in giving
the vehicle the green light in all trials, and on average the amount of
saved time due to TSP algorithm varied from 8.0 s or 32% (for
arriving almost at the end of the red light cycle of the intersection)
to 41.0 s or 75% (for arriving almost at the early red light cycle of
the intersection).
Comparisons
Overall, there were 14 matching cases among the trials with regular
GPS and those with differential GPS [cycle length start time 40 s
(three trials for each), 50 s (two trials for each), 60 s (three trials for
each), 70 s (three trials for each), and 80 s (three trials for each)].
Because there were different numbers of trials for some scenarios,
in order to perform the statistical test the study team randomly se-
lected an equal number of trials from each GPS type.
At first sight, regular GPS and differential GPS show no
significant difference because both types of GPS guaranteed a
100% successful pass rate. The research took one step further to
investigate the following factors to show the potential difference
in terms of operation with the two types of GPS devices:
•Actual overall time to pass the intersection;
•Reduced delay (in seconds and percentages); and
•Green extension duration.
Table 7summarizes these two values for both devices.
A paired t-test was conducted to statistically examine the effect
of differential GPS on average travel times, mean reduced delays
(in seconds and percentages), and mean green extension times of
the two devices. Table 8shows the results.
Based on Table 8,p-values for all tests are larger than 0.05. This
indicates that there is no significant difference between the two de-
vices. The results of the comparison for actual overall time to pass
Table 6. Differential GPS Scenario
Cycle
length
start
time
(s) Trial
Speed at
0.8 km
(0.5 mi)
[km=h (mi=h)] PRCAT
Predicted
original
traffic
light status
without
TSP
Predicted
overall
time to
pass the
intersection
without TSP
Actual
traffic
light
status
with
TSP
Actual
overall
time to
pass the
intersection
with TSP
Green
extension
(s)
Delay
without
TSP
(s)
Delay with
TSP
(s)
Reduced
delay
(s)
Reduced
delay
(%)
40 1 78.2 (48.6) 89.8 Red 105.0 Green 64.0 0 65.0 24.0 41.0 63
2 76 (47.2) 1.9 Red 105.0 Green 65.0 2 65.0 25.0 40.0 62
3 74.7 (46.4) 4.0 Red 105.0 Green 62.0 0 65.0 22.0 43.0 66
50 1 75.3 (46.8) 13.1 Red 95.0 Green 54.0 0 55.0 14.0 41.0 75
2 80.1 (49.8) 10.8 Red 95.0 Green 54.0 0 55.0 14.0 41.0 75
60 1 71.3 (44.3) 24.4 Red 85.0 Green 57.0 2 45.0 17.0 28.0 62
2 75.2 (46.7) 23.4 Red 85.0 Green 59.0 3 45.0 19.0 26.0 58
3 71.3 (44.3) 25.1 Red 85.0 Green 58.0 2 45.0 18.0 27.0 60
70 1 73.2 (45.5) 32.7 Red 75.0 Green 55.0 0 35.0 15.0 20.0 57
2 75.2 (46.7) 30.9 Red 75.0 Green 54.0 0 35.0 14.0 21.0 60
3 79.5 (49.4) 27.4 Red 75.0 Green 51.0 0 35.0 11.0 24.0 69
80 1 70.2 (43.6) 44.5 Red 65.0 Green 56.0 1 25.0 16.0 9.0 36
2 73.5 (45.7) 41.6 Red 65.0 Green 57.0 3 25.0 17.0 8.0 32
3 63.2 (39.3) 49.7 Red 65.0 Green 58.0 3 25.0 18.0 7.0 28
Table 7. Actual Overall Times and Green Extension Times for Regular GPS and Differential GPS
Cycle
length
start
time (s) Trial
Actual overall time to
pass the intersection
with TSP
Reduced
delay
(s)
Reduced
delay
(%)
Green
extension
(s)
Trial
Actual overall time to
pass the intersection
with TSP
Reduced
delay
(s)
Reduced
delay
(%)
Green
extension
(s)
Regular GPS Differential GPS
40 1 63.0 42.0 65 0 1 64.0 41.0 63 0
6 80.0 42.0 65 5 2 65.0 40.0 62 2
9 67.0 38.0 58 4 3 62.0 43.0 66 0
50 2 56.0 39.0 71 0 1 54.0 41.0 75 0
3 56.0 39.0 71 0 2 54.0 41.0 75 0
60 1 56.0 29.0 64 0 1 57.0 28.0 62 2
2 56.0 29.0 64 1 2 59.0 26.0 58 3
3 56.0 29.0 64 0 3 58.0 27.0 60 2
70 1 58.0 17.0 49 1 1 55.0 20.0 57 0
2 59.0 16.0 46 3 2 54.0 21.0 60 0
3 55.0 20.0 57 0 3 51.0 24.0 69 0
80 1 54.0 11.0 44 0 1 56.0 9.0 36 1
2 56.0 9.0 36 0 2 57.0 8.0 32 3
3 56.0 9.0 36 1 3 58.0 7.0 28 3
Mean —59.1 26.4 56 1.1 —57.4 26.9 57 1.1
© ASCE 05017005-10 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

the intersection reveal that in an aggregated sense, the difference in
overall time to pass the intersection is not statistically significant
(p¼0.099). There is also a high correlation between actual overall
time to pass the intersection of the two devices (Pearson corre-
lation = 0.752), which indicates the two devices operate similarly.
Table 8also shows the results for reduced delay in seconds and
percentage. The difference between the two devices is not statisti-
cally significant; moreover, high correlation between the values
of the regular GPS and differential GPS (Pearson correlation =
0.975 and 0.867, respectively) indicate the two devices perform
similarly.
The results of the comparison for green extension also indicate
that in an aggregate sense, the difference in green extension time is
not statistically significant (p¼0.452). Moreover, there is no cor-
relation between the green extension times for the two devices
(Pearson correlation ¼−0.04), which indicates the green extension
times for the two devices were random and not subject to the GPS
device.
This is an important finding that shows no statistically signifi-
cant difference between the operation of a regular GPS and
differential GPS for TSP. This finding can facilitate the large-scale
implementation of the TSP because the regular GPS devices are
much cheaper than differential GPS devices.
Conclusions
A field test on the previously developed TSPCV algorithm was
conducted at the Smart Road of the Virginia Tech Transportation
Institute in Blacksburg, Virginia. Although this experiment was
performed in an ideal condition without any other traffic, this study
is one of a few TSP studies that validate the TSP performance by
field experiment and one of the first TSP experiments with an in-
telligent TSP environment, and its results could help providing
guidance to future and similar field experiements.
Based on the CV-based TSP field experiment, the following
conclusions were made:
1. The TSPCV algorithm implementation program, data pack-
ets, and communications devices worked properly in a CV
environment.
2. The implementation of the TSPCV algorithm was successful for
the scenarios tested in this experiment (i.e., two GPS devices
and five sets of red light arrival times), and could reduce an aver-
age of 57% of bus delay at the intersection. The magnitude of
reduction in this study equals an average of 26.9 s at one inter-
section for a bus with speed of 72.4km=h(45 mi=h) and a traf-
fic signal with a 90-s cycle length with 30-s green time.
3. The TSPCV algorithm provided the bus green time at a 100%
success rate and it reduced delays between 32 and 75% (in com-
parison with no TSP). Delay reductions were observed to be
higher when the bus was predicted to meet the beginning of ori-
ginal red light without the TSPCV algorithm.
4. For approximately 50% of the experiment scenarios (both reg-
ular and differential GPS devices), green extension was needed
for the bus to pass the intersection during the TSPCV.
5. The performances of regular and differential GPS devices were
not statistically signficantly different at the 95th percentile con-
fidence level. This finding could help deploy large-scale imple-
mentation of the TSP because regular GPS devices are readily
available and much cheaper than differential GPS devices.
While the experimental results from this paper confirmed the
TSPCV works properly in a controlled environment, future re-
search in the real world with general traffic is necessary. It is likely
that the 100% TSP success rate would not always be possible if
queues and unexpected delays happen at the intersections. Thus,
it is recommended that the TSPCV be tested in real-world condi-
tions to ensure the TSPCV performs adequately in various traffic
conditions. In addition, the test is conducted in a suburban environ-
ment [0.8 km (0.5 mi) intersection spacing] because of the limita-
tions of the developed controller. Since the experiment, the research
team has enhanced the controller for more urban and tight spac-
ing, so future work could consider extending evaluation to urban
environment.
Acknowledgments
This research was in part supported by the Connected Vehicle/
Infrastructure University Transportation Center, and the GRL
Program through the National Research Foundation of Korea
(NRF) funded by the Ministry of Science, ICT & Future Plan-
ning (2013K1A1A2A02078326). The research team appreciates
Melissa Hulse, Leslie Harwood, Gabrielle Laskey, Tammy Russel,
Andy Peterson, Jean Paul Talledo Vilela, and Zac Doerzaph at the
Virginia Tech Transportation Institute and Andrew Farkas, Anita
Jones, and Nancy Jackson at the National Transportation Center
at Morgan State University for their support. The research team also
appreciates Kaveh Bakhsh Kelarestaghi for the data collection.
References
Ahn, K., Rakha, H., and Hale, D. K. (2015). “Multi-modal intelligent traffic
signal systems (MMITSS) impacts assessment.”Final Rep., U.S. Dept.
of Transportation, Washington, DC.
Table 8. t-Test Results between Regular GPS and Differential GPS
Statistic
Actual overall time Reduced delay (s) Reduced delay (%) Green extension
Regular
GPS
Differential
GPS
Regular
GPS
Differential
GPS
Regular
GPS
Differential
GPS
Regular
GPS
Differential
GPS
Mean 59.14 57.43 26.36 26.86 0.56 0.57 1.07 1.14
Variance 48.13 16.11 156.86 168.75 0.015 0.022 2.84 1.67
Observations 14 14 14 14 14 14 14 14
Pearson correlation 0.75 —0.98 —0.87 —−0.040 —
Hypothesized mean difference 0 —0—0—0—
Degrees of freedom 13 —13 —13 —13 —
t-statistic 1.36 —−0.65 —−0.41 —−0.12 —
PðT≤tÞone-tail 0.099 —0.26 —0.35 —0.45 —
tcritical one-tail 1.77 —1.77 —1.77 —1.77 —
PðT≤tÞtwo-tail 0.20 —0.53 —0.69 —0.90 —
tcritical two-tail 2.16 —2.16 —2.16 —2.16 —
© ASCE 05017005-11 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.

Al-Sahili, K., and Taylor, W. (1996). “Evaluation of bus priority signal
strategies in Ann Arbor, Michigan.”Transp. Res. Rec., 1554, 74–79.
Balke, K., Dudek, C., and Urbanik, T. (2000). “Development and evaluation
of intelligent bus priority concept.”Transp. Res. Rec., 1727, 12–19.
Chang, G. L., Vasudevan, M., and Su, C. C. (1996). “Modeling and evalu-
ation of adaptive bus-preemption control with and without automatic
vehicle location systems.”Transp. Res. Part A: Policy Pract., 30(4),
251–268.
Ding, J., He, Q., Head, L., Saleem, F., and Wu, W. (2013). “Develop-
ment and testing of priority control system in connected vehicle
environment.”92nd Annual Meeting, Transportation Research Board,
Washington, DC.
Ekeila, W., Sayed, T., and El Esawey, M. (2009). “Development of dynamic
transit signal priority strategy.”Transp. Res. Rec., 2111, 1–9.
Garrow, M., and Machemehl, R. B. (1997). “Development and evaluation
of transit signal priority strategies.”Research Rep., Southwest Region
Univ. Transportation Center, Texas Transportation Institute, College
Station, TX.
He, Q., Head, K., and Ding, J. (2011). “Heuristic algorithm for priority
traffic signal control.”Transp. Res. Rec., 2259, 1–7.
He, Q., Head, K. L., and Ding, J. (2014). “Multi-modal traffic signal control
with priority, signal actuation and coordination.”Transp. Res. Part C:
Emerging Technol., 46, 65–82.
Hu, J., Park, B., and Parkany, A. (2014). “Transit signal priority with con-
nected vehicle technology.”Transp. Res. Rec., 2418, 20–29.
Hu, J., Park, B. B., and Lee, Y. J. (2015). “Coordinated transit signal prior-
ity supporting transit progression under connected vehicle technology.”
Transp. Res. Part C: Emerging Technol., 55, 393–408.
Khasnabis, S., Karnati, R., and Rudraraju, R., (1996). “NETSIM-based ap-
proach to evaluation of bus preemption strategies.”Transp. Res. Rec.,
1554, 80–89.
Kimpel, T., Strathman, J., Bertini, R., and Callas, S. (2005). “Analysis of
transit signal priority using archived TriMet bus dispatch system data.”
Transp. Res. Rec., 1925, 156–166.
Lee, J., Shalaby, A., Greenough, J., Bowie, M., and Hung, S. (2005).
“Advanced transit signal priority control with online micro-simulation-
based transit prediction model.”Transp. Res. Rec., 1925, 185–194.
Liao, C., and Davis, G. A. (2007). “Simulation study of a bus signal priority
strategy based on GPS, AVL and wireless communications.”86th
Annual Meeting, Transportation Research Board, Washington, DC.
Liao, C. F., and Davis, G. A. (2011). “Field testing and evaluation of a
wireless-based transit signal priority system.”Final Rep., Intelligent
Transportation Systems Institute, Center for Transportation Studies,
Univ. of Minnesota, Minneapolis.
Ma, W., Yang, X., and Liu, Y. (2010). “Development and evaluation of a
coordinated and conditional bus priority approach.”Transp. Res. Rec.,
2145, 49–58.
Muthuswamy, S., McShane, W., and Daniel, J. (2007). “Evaluation of
transit signal priority and optimal signal timing plans on transit and
traffic operations.”86th Annual Meeting, Transportation Research
Board, Washington, DC.
Oliveira-Neto, F., Loureiro, C., and Han, L. (2009). “Active and passive
bus priority strategies in mixed traffic arterials controlled by SCOOT
adaptive signal system assessment of performance in Fortaleza, Brazil.”
Transp. Res. Rec., 2128, 58–65.
Pratt, R. H., et al. (2000). Traveler response to transportation system
changes: Interim handbook, Transportation Research Board, National
Resource Council, Washington, DC.
Rakha, H., and Ahn, K. (2006). “Transit signal priority project—Phase II.”
Virginia Transportation Research Council, Charlottesville, VA.
University of Arizona, University of California PATH Program, Savari
Networks, Inc., Econolite. (2016). “Multi-modal intelligent traffic
signal system—Phase II: System development, deployment and field
test.”Final Rep., Tucson, AZ.
Wang, Y., Hallenbeck, M., Zheng, J., Zhang, G., Ma, X., and Corey, J.
(2008). “Comprehensive evaluation of transit signal priority system
impacts using field observed traffic data.”Final Technical Rep.,
Washington State Dept. of Transportation, Olympia, WA.
Wang, Y., Ma, W., Yin, W., and Yang, X. (2014). “Implementation
and testing of cooperative bus priority system in connected vehicle envi-
ronment case study in Taicang City, China.”Transp. Res. Rec., 2424,
48–57.
Wang, Y., Yang, X., Huang, L., and Zhang, L. (2013). “A phase-by-phase
traffic control policy at isolated intersection based on cooperative
vehicle-infrastructure system.”13th COTA Int. Conf. of Transporta-
tion Professionals (CICTP 2013), Vol. 96, Procedia—Social and
Behavioral Sciences, 1987–1996.
Yelchuru, B., et al. (2014). “AERIS—Applications for the environment:
Real-time information synthesis, eco-signal operations operational
scenario modeling report.”ITS Joint Program Office-HOIT, U.S.
Dept. of Transportation, Washington, DC.
© ASCE 05017005-12 J. Transp. Eng., Part A: Syst.
J. Transp. Eng., Part A: Systems, 2017, 143(8): -1--1
Downloaded from ascelibrary.org by University of Virginia on 05/06/17. Copyright ASCE. For personal use only; all rights reserved.