ISSN 1055- 1425
May 2010
This work was performed as part of the California PATH Program of the
University of California, in cooperation with the State of California Business,
Transportation, and Housing Agency, Department of Transportation, and the
United States Department of Transportation, Federal Highway Administration.
The contents of this report reflect the views of the authors who are responsible
for the facts and the accuracy of the data presented herein. The contents do not
necessarily reflect the official views or policies of the State of California. This
report does not constitute a standard, specification, or regulation.
Final Report for Task Order 6407
CALIFORNIA PATH PROGRAM
INSTITUTE OF TRANSPORTATION STUDIES
UNIVERSITY OF CALIFORNIA, BERKELEY
Relieve Congestion and Conflicts Between
Railroad and Light Rail Grade- Crossing
Intersections
UCB- ITS- PRR- 2010- 29
California PATH Research Report
Wei- Bin Zhang et. al
CALIFORNIA PARTNERS FOR ADVANCED TRANSIT AND HIGHWAYS
i
Task Order 6407 ( in continuation of TO5407)
Relieve Congestion and Conflicts Between
Railroad and Light Rail Grade- Crossing
Intersections
Prepared by:
California PATH
University of California, Berkeley
and
California Department of Transportation
in collaboration with
SANDAG, San Diego Trolley, Inc. ( SDTI), City of San Diego
Final Report for TO 6407
ii
Final Report for TO 6407
iii
ACKNOWLEDGMENTS
This project is sponsored by the California Department of Transportation ( Caltrans)
under Task Order 6407 with Task number 0742 under Project P567 " Transit Rail Right of
Way Safety. This report was prepared in cooperation with the State of California,
Business Transportation and Housing Agency, Department of Transportation, San Diego
Association of Government ( SANDAG), San Diego Trolley, and City of San Diego. The
contents of this report reflect the views of the authors, who are responsible for the facts
and the accuracy of the data presented herein. The contents do not necessarily reflect the
official views or policies of the State of California. This report does not constitute a
standard, specification, or regulation.
The authors of this report would also like to express our appreciation to Dan Lovegren of
Caltrans, Samuel Johnson, Steve Celniker, Chiachi Rumbolo and Alex Estrella of
SANDAG for their guidance and support; Yue Li, Scott Johnston, and Susan Dickey of
California PATH; Duncan Hughes and Eddie Flores of City of San Diego; Steve Brown
of McCain; and Don Murphy of IBI Group for their technical assistance and support.
Author List
University of California, Berkeley:
Wei- Bin Zhang ( Principal Investigator)
Meng Li
Guoyuan Wu
Kun Zhou
Fanping Bu
Final Report for TO 6407
iv
Final Report for TO 6407
v
EXECUTIVE SUMMARY
This report specifically summarizes the work that the PATH team has performed at this
stage of Task Order 6407 in continuation of previous Task Order 5407. We have
conducted an in- depth study of problems associated with grade crossings for this project.
We started from the system design based on the proposed adaptive trolley signal priority
( ATSP) system. The system is designed for large- scale field implementation of the ATSP
system. It consists of three sub- systems: onboard sub- system, roadside control sub-system
and central control sub- system. The system input and output diagram is based on
the previous developed algorithm and programs.
We designed and conducted the laboratory testing for the ATSP system. The laboratory
setting is the step prior to field operational testing of the proposed ATSP system in San
Diego. The objective of the laboratory testing is to testify and demonstrate the
applicability of the proposed system, particularly the communication system and the
traffic signal operation system in San Diego, i. e. the QuicNet/ 4 central control system in
the Transportation Management Center ( TMC) and Type 170 controllers at roadside
running McCain’s Bitran 233 control software. There are two steps in the laboratory
testing. The first step is to show the proposed system in an entirely closed laboratory
environment. The second step is to move the testing one step closer to the FOT
environment and involves field signal operation systems and the actual communication
system.
The preliminary FOT was designed and conducted with the objective to demonstrate
proof- of- concept of the proposed adaptive trolley signal priority ( ATSP) in San Diego
and to evaluate the potential applicability for such a system in a large- scale
implementation.
Based on discussion with the City of San Diego and SANDAG, the testing site selected
was the 0.8- mile- long arterial segment of C Street in Downtown San Diego. There are
four trolley stations along this site. From the West side, they are America Plaza, Civic
Center, 5th Ave., and City College. The site consists of fifteen signalized intersections
from India St. to 10th Ave. Two trolley lines are serving this segment of C Street. They
are the Blue and Orange Lines with regular service headway fifteen minutes. During the
peak hour, the Blue Line runs higher frequent service with headway seven minutes.
There are two stages of data collection for the FOT. Stage one is for the “ before” scenario
in which trolleys do not experience any signal priority. Stage one is from October 30,
2009 to November 8, 2009. Stage two is the “ after” scenario in which selected trolley
trains are able to request transit signal priority along the testing corridor. Stage two
started on October 16, 2009 and ended on October 26, 2009.
A thorough data analysis was conducted after the preliminary FOT. A successful trip was
presented and analyzed. The proposed ATSP system was able to significantly reduce the
Final Report for TO 6407
vi
Final Report for TO 6407
vii
number of stops and stop times for the trolley trip. However, the overall performance of
the proposed ATSP system was not as successful as expected. The maximum average
reduction on average number of stops and average travel time is less than 15%. More
issues were observed and studied from the perspective of trolley operation, traffic
operation, and prediction for trolley movement and station dwelling time.
At the end of the report, all the issues were summarized. The project team also provides
recommendations in order to further improve the system towards the next research step.
The recommendations cover six aspects: signal transition, signal progression, dwelling
time prediction, arrival time prediction at stations, integration of priority decision with
prediction, and the automatic vehicle location ( AVL) system. With all the system
improvements and testing, the final FOT as the next step will be performed.
It is also noted that the proposed system design can be easily applied to other light- rail
transit ( LRT) systems, which do not have the preempted right- of- way at grade crossings
and intersections. It is not necessary to have fixed- timing control at the signalized
intersections. The proposed concept and system can be easily adapted to actuated or
semi- actuated control systems. Moreover, the results and lessons learned from both of the
laboratory test and the preliminary field operational test could help in designing and
calibrating such transit signal priority systems for other similar LRT systems.
Final Report for TO 6407
viii
Final Report for TO 6407
ix
Table of Contents
TABLE OF CONTENTS....................................................................................................................... .... IX LIST OF FIGURES ............................................................................................................................... .... XI LIST OF TABLES ............................................................................................................................... ... XIII 1. SYSTEM DESIGN......................................................................................................................... ....... 1 2. LABORATORY TESTING .................................................................................................................. 4 2.1 TESTING PURPOSE .............................................................................................................................. 4 2.2 TESTING DESIGN ............................................................................................................................... 4 2.3 LABORATORY TESTING AT PT2L........................................................................................................ 4 2.4 LABORATORY TESTING AT SAN DIEGO TMC..................................................................................... 5 3. PRELIMINARY FIELD OPERATIONAL TEST.............................................................................. 9 3.1 TESTING PURPOSE .............................................................................................................................. 9 3.2 TESTING DESCRIPTION ....................................................................................................................... 9 3.3 RESULTS ANALYSIS ......................................................................................................................... 11 3.3.1 System Performance ................................................................................................................ 11 3.3.2 Prediction Analysis.................................................................................................................. 20 4. RECOMMENDATIONS AND NEXT STEP .................................................................................... 27 4.1 SIGNAL TRANSITION ........................................................................................................................ 27 4.2 SIGNAL PROGRESSION.................................................................................................................... . 28 4.3 DWELL TIME PREDICTION................................................................................................................ 29 4.4 TROLLEY’S ARRIVAL TIME PREDICTION AT STATION...................................................................... 29 4.5 INTEGRATING PRIORITY DECISION WITH PREDICTION ..................................................................... 30 4.6 AUTOMATIC VEHICLE LOCATION ( AVL) SYSTEM........................................................................... 32 5. APPENDIX ............................................................................................................................... ........... 33
Final Report for TO 6407
x
Final Report for TO 6407
xi
List of Figures
Figure 1- 1 Physical architecture of San Diego ATSP System............................................ 1 Figure 1- 2 Cell Phone based AVL system.......................................................................... 2 Figure 1- 3 Input output diagram for the priority request generator.................................... 3 Figure 2- 1 One typical southbound/ outbound laboratory testing trip................................. 7 Figure 3- 1 Map of testing Site ............................................................................................ 9 Figure 3- 2 Trajectory of an example trip from Civic Center to 5th Ave........................... 13 Figure 3- 3 Changes on phase 2 at 8th Ave ........................................................................ 18 Figure 3- 4 Changes on phase 2 at India Street ................................................................. 18 Figure 3- 5 Changes on phase 4 at India Street ................................................................. 20 Figure 3- 6 GPS trajectories for two Orange Line trips..................................................... 21 Figure 3- 7 GPS trajectories for two Blue Line trips ......................................................... 21 Figure 3- 8 A typical trolley trajectory .............................................................................. 22 Figure 3- 9 Distribution of prediction errors without stopping time at stations ................ 23 Figure 3- 10 Distribution of prediction errors with stopping time at stations ................... 23 Figure 3- 11 Distribution of dwelling time at America Plaza Station ............................... 24 Figure 3- 12 Distribution of dwelling time at 5th Ave Station ........................................... 25 Figure 4- 1 Outbound trajectories between America Plaza and Civic Center ( No TSP)... 28 Figure 4- 2 Inbound trajectories between Civic Center and America Plaza ( No TSP) ..... 29 Figure 4- 3 Trolley stop time at City College Station ( No TSP) ....................................... 31 Figure 4- 4 Green bands along testing segment ( No TSP) ................................................ 31 Figure 4- 5 GPS receptions with GPS external antenna .................................................... 32 Figure 5- 1 Changes on phase 2 ( trolley) at Front Street................................................... 34 Figure 5- 2 Changes on phase 2 ( trolley) at 5th Street ....................................................... 34 Figure 5- 3 Changes on phase 2 ( trolley) at 6th Street ....................................................... 35 Figure 5- 4 Changes on phase 2 ( trolley) at 7th Street ....................................................... 35 Figure 5- 5 Changes on phase 2 ( trolley) at 8th Street ....................................................... 36 Figure 5- 6 Changes on phase 2 ( trolley) at 10th Street ..................................................... 36 Figure 5- 7 Changes on phase 2 ( trolley) at 11th Street ..................................................... 37 Figure 5- 8 Changes on phase 4 ( trolley) at Front Street................................................... 37 Figure 5- 9 Changes on phase 4 ( trolley) at 5th Street ....................................................... 38 Figure 5- 10 Changes on phase 4 ( trolley) at 6th Street ..................................................... 38 Figure 5- 11 Changes on phase 4 ( trolley) at 7th Street ..................................................... 39 Figure 5- 12 Changes on phase 4 ( trolley) at 8th Street ..................................................... 39 Figure 5- 13 Changes on phase 4 ( trolley) at 10th Street ................................................... 40 Figure 5- 14 Changes on phase 4 ( trolley) at 11th Street ................................................... 40 Figure 5- 15 Outbound trajectories between Civic Center and 5th Ave ( No TSP) ............ 41 Figure 5- 16 Outbound trajectories between 5th Ave and City College ( No TSP) ............ 41 Figure 5- 17 Inbound trajectories between City College and 5th Ave ( No TSP) ............... 42 Figure 5- 18 Inbound trajectories between 5th Ave and Civic Center ( No TSP) ............... 42
Final Report for TO 6407
xii
Final Report for TO 6407
xiii
List of Tables
Table 2- 1 Sensitivity analysis for 3rd Ave and 4th Ave ....................................................... 5 Table 2- 2 Sensitivity analysis for 5th Ave and the section.................................................. 5 Table 3- 1 Summary of trip samples.................................................................................. 10 Table 3- 2 Detailed trip samples for Stage 2...................................................................... 10 Table 3- 3 Summary of execution rates for requests ......................................................... 12 Table 3- 4 Performance of an example trip from Civic Center to 5th Ave ........................ 14
Table 3- 5 Original and proposed timings for the example trip......................................... 14 Table 3- 6 Summary of number of stops at signals ........................................................... 15 Table 3- 7 Request non- blockage rates.............................................................................. 16 Table 3- 8 Summary of impacts on signal cycles .............................................................. 16 Table 3- 9 Summary of changes on phase 2 ( trolley phase) .............................................. 17 Table 3- 10 Summary of changes on phase 4 ( general traffic) .......................................... 19 Table 5- 1 Detailed trip samples for Stage 1 without TSP ................................................ 33
Final Report for TO 6407
xiv
Final Report for TO 6407
1
1. System Design
As illustrated in Figure 1- 1, the field- testing system consists of three sub- systems:
onboard sub- system, roadside control sub- system and central control sub- system. PATH
has developed a cost- effective solution for automatic vehicle location ( AVL) systems, as
shown in Figure 1- 2. This system is based on Motorola iDEN phones with built- in GPS
receivers and Java platform micro edition ( J2ME). Although it is proved that the phone-based
AVL system is sufficient to support adaptive transit signal priority along major
arterials in the San Francisco Bay Area, it is still uncertain that such a system would be
appropriate for the adaptive trolley priority system in downtown San Diego. As one of
the objectives for the FOT, it is also to demonstrate the applicability of such cost-effective
AVL systems to support ATSP in San Diego. On nine trolley trains, the cell
phone- based system has been installed either on the roof of the operator’s room or behind
the destination sign window in front of the train depending on the power availabilities.
Figure 1- 1 Physical architecture of San Diego ATSP System
Final Report for TO 6407
2
Figure 1- 2 Cell Phone based AVL system
For the roadside sub- system, no additional equipment is installed in the controller cabinet
for FOT testing. The existing two- way communication link between each local signal
controller to the central traffic management center ( TMC) has been examined before the
FOT. In one direction, the signal controller is able to send detailed traffic operation
information, e. g. current phase, running pattern, local clock timer, etc., in real time. In the
opposite direction, the TMC can send priority requests with a set of proposed force- off
points to the designated intersection. Once the request is received at the local signal
controller, the signal timing would be changed before the next start of a signal cycle for
the implementation of an ATSP request.
The central control sub- system consists of the QuicNet/ 4 server computer, the priority
request generator computer, and the Ethernet communication link between them. The
QuicNet/ 4 server manages communication between the central system and the local
signal controllers in the field. Through the Ethernet communication link and an interface
program jointly developed by McCain and PATH, the priority request generator
computer receives the real- time traffic operation information from the local controllers in
the field. Together with the real- time trolley location information from the AVL systems
on trolleys, the offline optimization module is able to select the optimal force- off points
for the intersections to provide priority to trolley trains when they arrive. Figure 1- 3
illustrates the input and output diagram for the priority request generator. The
optimization module runs in a Linux operating system and a data hub message queuing
environment to support real- time operation.
Final Report for TO 6407
3
Figure 1- 3 Input output diagram for the priority request generator
Final Report for TO 6407
4
2. Laboratory Testing
2.1 Testing Purpose
The laboratory is the step prior to field operational testing of the proposed ATSP system
in San Diego. The objective of the laboratory testing is to testify and demonstrate the
applicability of the proposed system, particularly the communication system and the
traffic signal operation system in San Diego, i. e. the QuicNet/ 4 central control system in
the TMC and Type 170 controllers at roadside running McCain’s Bitran 233 control
software.
2.2 Testing Design
There are two steps in the laboratory testing. The first step is to show the proposed
system in an entirely closed laboratory environment. The second step is to move the
testing one step closer to the FOT and involves field signal operation systems and the
actual communication system.
2.3 Laboratory Testing at PT2L
McCain and PATH set up the testing environment at Parsons Traffic and Transit
Laboratory ( PT2L) at PATH. The testing platform consists of three Type 170 signal
controllers with McCain’s Bitran 233 programs, a server computer with McCain’s
QuicNet/ 4 software installed, the communication links between the three signal
controllers and the QuicNet/ 4 server computer, a PATH control computer with all the
TSP software installed, and the communication link between the QuicNet/ 4 server and
the PATH control computer. PATH then debugged and tested all the TSP software in this
testing environment at PATH.
The original configurations of the signal controller setting in PATH’s PT2L are not
identical with those at the San Diego Traffic Management Center ( TMC). In particular,
the “ pre- timed” operation for phase 4 was enabled on the Bitrans 233 program. When
McCain set up the controllers at PT2L, the remote communication to SD’s TMC had not
been established yet. Thus McCain was not able to testify the settings with the field
controllers. Under the incorrect settings at PT2L, the force- off point of phase 4 is the time
when the yellow of phase 4 starts. Under the field settings, phase 4 should be force- off at
the beginning of its flash- don’t- walk period. Under the help from the City of San Diego
and McCain, such settings have been corrected. The control logic has been extensively
examined. In addition, new constraints on force- off points of both phases as well as
permissive end have also been tested in detail.
Final Report for TO 6407
5
Because of the rectification of the signal controller settings, constraints of some
parameters, e. g. force- off point of phase 4, or the relationship among parameters, e. g. the
gap between force- off point of phase 2 and phase 4, were modified accordingly.
According to the results, the feasible region of the proposed optimization model is a bit
smaller than that without the modification.
In the lab testing, if there is no disturbance on the prediction departure/ arrival time at
stations/ signals, then all trips for both directions will experience zero- stop along all
intersections between stations. However, most situations in the real world are far from
being ideal and the prediction results cannot be guaranteed to be perfect at all. Therefore,
a sensitivity analysis on the prediction error is indispensable. The results are shown in
Table 2- 1 and Table 2- 2.
Table 2- 1 Sensitivity analysis for 3rd Ave and 4th Ave
Delay at 3rd Ave Sample Trips Delay at 4th Ave
Mean STD Mean STD
STD= 0 29 0.17 0.38 0 0.00
STD= 2 29 0.45 1.50 0.48 2.60
STD= 5 28 6.89 16.92 5.11 15.18
STD= 9 28 3.04 11.07 4.68 11.44
STD= 14 27 5.48 10.79 13.59 17.09
STD= 20 27 3.78 10.44 11.00 18.36
As can be observed from the results, system performance becomes worse as the standard
deviation of the prediction error ( unbiased prediction is assumed) gets larger. In other
words, the overall delay of the simulated section, on average, keeps increasing. At the
same time, the variation of such delay becomes more and more noticeable.
Table 2- 2 Sensitivity analysis for 5th Ave and the section
Sample Delay at 5th Ave Section Delay
Trips Mean STD Mean STD
STD= 0 29 0 0 0.17 0.38
STD= 2 29 2.28 11.11 3.21 11.51
STD= 5 28 0.25 0.80 12.25 21.00
STD= 9 28 3.86 10.34 11.57 21.68
STD= 14 27 4.45 12.39 23.52 21.32
STD= 20 27 4.59 13.24 19.37 23.29
2.4 Laboratory Testing at San Diego TMC
PATH worked with the City of San Diego and McCain and set up the testing
environment at the San Diego TMC. The testing platform was quite similar with the one
at PATH. It consisted of five Type 170 signal controllers with McCain’s Bitran 233
programs, the TMC QuicNet/ 4 server computer with McCain’s communication software
installed, the communication links between the five signal controllers and the QuicNet/ 4
Final Report for TO 6407
6
server computer, a PATH control computer with all the TSP software installed, and the
communication links between QuicNet/ 4 server and the PATH control computer and
between the PATH control computer and the PATH server at PT2L. PATH and the IT
group at the City of San Diego set up a reverse connection so that the PATH control
computer can receive the trolley GPS data from the PATH server computer in Berkeley.
All the hardware and communication links were tested at the San Diego TMC. The five
controllers were set up with identical settings as five field intersections at C Street: India
Street, 3rd Avenue, 4th Avenue, 5th Avenue, and 6th Avenue.
Before using the trolley GPS data from the field to test our system, we first conducted the
lab test with simulated trolley runs on the TMC testing platform. Under this scenario, our
trolley simulation tool generated trolley trips and mimicked train movements. We set up
the study corridor with 13 signalized intersections and sent out one trolley to travel back
and forth. During the lab testing, five of the 13 traffic signals were controlled by real
signal controllers as described above. The trolleys’ historical movement data served as
input parameters of our optimization model. In addition, the dwelling times at those four
relevant stations, i. e. American Plaza, Civic Center, 5th Avenue and City College, came
from both the historical operation data collected by the PATH automatic vehicle location
( AVL) system and some latest field surveys conducted in June 2008. Starting from June
18th 2009, we have run the lab test in the San Diego TMC continuously for four days.
With simulated trolley runs, we obtained over 500 trolley runs with equal number of trips
for both Southbound/ Outbound and Northbound/ Inbound directions. Given the perfect
prediction of train movements and dwell times, all trolley runs under signal priority were
able to travel through signalized intersections without any unnecessary stops ( i. e. non-station
stops), except for those trips released at around midnight. The cause of the stops
was due to abnormal controller operation, which will be described in detail in a later
section. To analyze the simulation data, we developed our tools using MATLAB to
visualize the results and get further insight of the trolley and signal operations. Figure 2- 1
shows one typical Southbound/ Outbound trip in the lab test. After detecting the incoming
trolley, the PATH control computer generated signal priority requests based on the
movement of trolley trains and signal timings from the QuicNet/ 4 server, which then
downloaded the signal timings onto the three controllers that were set up in the TMC.
After the controllers implemented the new timings, the simulated trolley with priority
was able to go through all three intersections without any stops. Note that the dwell times
have been equivalently converted to travel times.
Final Report for TO 6407
7
Figure 2- 1 One typical southbound/ outbound laboratory testing trip
Based on our observations and discussions with City of San Diego engineers, we learned
that all signal controllers along the study corridor are reset at approximately midnight
( 00: 00 A. M.) each day. During this time, all signal controllers are in transition for at least
150 seconds. This caused the trolley request to be dropped or not properly implemented.
As a tentative solution, we can block out a 10- minute period around midnight when we
would not process any signal priority requests.
Although our prediction tool is trying to filter out the GPS- related error, the worst of GPS
receptions will result in the worst of prediction outputs. Subsequently, the traffic signal
timings can hardly be adjusted to adapt to the trolleys’ field movements. As mentioned
before, there are two types of GPS problems: GPS reception error and GPS signal loss.
Both of these errors are partially due to “ urban canyon” effects and the limitation of our
GPS devices. According to the test results, the quality of GPS data is adequate for the
field test with the purpose of verifying our ATSP system. For the field deployment of the
system, a more robust device will be needed.
The prediction of train dwell time is very difficult particularly with the random arrivals of
disabled people. However, this quantity is also a key parameter to our system because
signal controllers in pre- time mode require long lead- time to process timing change
requests. Based on extensive tests in the simulation environment, our algorithm will
definitely work well if the prediction is good enough. In comparison with trolleys’ travel
Final Report for TO 6407
8
time, dwell time is less consistent and more unpredictable. For example, if there is a
handicapped person who needs to board the train, the dwell time will get much longer
than usual. According to the field data analysis, trolleys’ waiting times ( may include
dwell time and signal waiting time) at the station can range from approximately 30
seconds to 3 minutes. One possible way to increase the accuracy of the dwell time
prediction is to build learning intelligence in the prediction software so that the prediction
tool can improve itself by learning from the collected field data.
Final Report for TO 6407
9
3. Preliminary Field Operational Test
3.1 Testing Purpose
The objective of the preliminary field operational test ( FOT) was to demonstrate proof-of-
concept for the proposed adaptive trolley signal priority ( ATSP) in San Diego and
evaluate the potential applicability for such a system in a large- scale implementation.
3.2 Testing Description
Based on discussions with the City of San Diego and SANDAG, the selected testing site
was the 0.8- mile- long arterial segment of C Street in Downtown San Diego, as shown in
Figure 3- 1 with four trolley stations along this site: From the west to east, they are
America Plaza, Civic Center, 5th Ave., and City College. The site consists of fifteen
signalized intersections from India St. to 10th Ave. and two trolley lines serve this
segment of C Street. They are the Blue and Orange Lines with regular service headway of
fifteen minutes. During the peak hour, the Blue Line runs more frequent service with
seven- minute headways.
Figure 3- 1 Map of testing Site
There were two stages of data collection for the FOT. Stage one was for the “ before”
scenario in which trolleys did not experience any signal priority. Stage one was from
October 30, 2009 to November 8, 2009. Stage two was “ after” scenario in which selected
trolley trains were able to request transit signal priority along the testing corridor. Stage
two started on October 16, 2009 and ended on October 26, 2009. Table 3- 1 presents the
summary of sample trips in the FOT. Table 5- 1 in the Appendix and Table 3- 2 illustrate
the detailed description of all the trip samples for Stage 1 and Stage 2, respectively.
Final Report for TO 6407
10
Table 3- 1 Summary of trip samples
Stage Number of trips
Outbound Inbound
1 ( No TSP) 67 79
2 ( With TSP) 109 123
Table 3- 2 Detailed trip samples for Stage 2
Date Trolley # 1 Trolley # 2 Trolley # 6 Trolley # 8 Summary
( Oct.)
OB* IB** OB IB OB IB OB IB OB IB
16th 0 0 5 5 1 2 3 3 9 10
17th 0 0 8 8 0 0 3 3 11 11
18th 0 0 0 0 0 0 5 4 5 4
19th 0 0 8 9 2 2 8 9 18 20
20th 0 0 7 8 2 2 1 1 10 11
21st 1 1 4 5 1 2 5 6 11 14
22nd 8 9 5 7 2 4 3 2 18 22
23rd 9 9 8 9 3 3 6 7 26 28
24th 7 7 2 1 0 0 3 3 12 11
25th 8 8 0 0 0 0 0 0 8 8
Sum. 33 34 47 52 11 15 37 38 128 139
* – Outbound trips include those operating along both Blue and Orange Lines
within the study scope
** – Inbound trips include those operating along both Blue and Orange Lines
within the study scope
The traffic signal timings serve as major inputs of the proposed optimization model.
Since last time PATH did the data collection, San Diego city engineers have updated
signal timings a few times. In order to prepare for the FOT, the most recent signal timing
sheets have been collected from the City of San Diego. All timing parameters in our
control software have been updated. In comparison with the previous version of traffic
signal timings, the changes include offsets, force- off points of phase 4, yellow intervals
and all red clearances.
Final Report for TO 6407
11
Due to the construction project around San Diego City College, the City College trolley
station was placed between 10th and 11th Avenues at C Street. Therefore, the original
study corridor was from India Avenue @ C Street to 10th Avenue @ C Street. The whole
corridor consisted of 12 signalized intersections. Upon the completion of the construction
project, the City College station was relocated between 11th Avenue @ C Street and Park
Boulevard @ C Street. Now there are 13 signalized intersections along the study corridor:
India Street, Front Street, 1st Avenue, 2nd Avenue, 3rd Avenue, 4th Avenue, 5th Avenue, 6th
Avenue, 7th Avenue, 8th Avenue, 9th Avenue, 10th Avenue and 11th Avenue. For the
additional intersection 11th Avenue @ C Street, we have collected and analyzed the
traffic signal timing information, geometry information, and traffic demand information.
11th Avenue @ C Street is quite unique from a geometric perspective because it has a
separate traffic phase, which parallels the trolley direction of movement. Therefore, we
made a few changes in our signal timing optimization software and generated the
prioritized signal timings.
In our previous work, we focused on signal timing Plan 2 for our study corridor. Plan 2
covers the time of day between 03: 00 and 15: 00. It is also consistent with the study
period that we set in the microscopic simulation model using PARAMICS. However, the
trolley operational span is longer than the time window mentioned above. The optimal
timing tables under Plan 4 are thus required and have been obtained by running the
proposed optimization model.
3.3 Results Analysis
3.3.1 System Performance
A successful implementation of the ATSP system depends on whether a priority request
can be properly generated, then communicated and finally deployed. Table 3- 3 presents
the execution rates for the priority requests at all the intersections along the test site. It is
observed that the majority of priority requests have been successfully executed. At most
signals, over 98% of requests can be successfully generated, communicated, and executed
at local signal controllers. At 11th Ave, there were five failure calls, which is 6% of all
requests. According to the communication log file, communication issues between the
QuicNet/ 4 server and the local signal controller likely caused the non- executions, e. g. at
11th Ave.
Final Report for TO 6407
12
Table 3- 3 Summary of execution rates for requests
Intersection Total Number of
Requests
Updated
Calls
Effective
Calls
Successful
Calls
Failure
Calls Successful Rate
India St 175 70 105 103 2 98%
Front St 181 85 96 96 0 100%
1st Ave 179 79 100 100 0 100%
2nd Ave 178 82 96 96 0 100%
3rd Ave 148 62 86 86 0 100%
4th Ave 145 53 92 91 1 99%
5th Ave 154 63 91 90 1 99%
6th Ave 149 60 89 89 0 100%
7th Ave 148 61 87 87 0 100%
8th Ave 145 59 86 86 0 100%
9th Ave 152 58 94 93 1 99%
10th Ave 143 56 87 87 0 100%
11th Ave 141 56 85 80 5 94%
3.3.1.1 Impacts on Trolley Operation
For part of the proof- of- concept for the proposed methodology, a real- world example trip
was taken to evaluate system performance. Figure 3- 2 shows the trajectory of the
example trip from the Civic Center Station to the 5th Ave Station. As is illustrated in the
figure, there is no stop on red along the three signalized intersections of 3rd Ave, 4th Ave
and 5th Ave between two stations, due to the successful execution of the priority request.
Final Report for TO 6407
13
Figure 3- 2 Trajectory of an example trip from Civic Center to 5th Ave
By carefully examining this trip, it can be observed that the actual departure time from
the Civic Center Station is 07: 09: 46 a. m., while the predicted departure time is 07: 09: 43
( only 3 seconds earlier). Based on such a good prediction, a priority request on the
changes of signal timings is generated and executed. As a result, the differences between
actual departure times and predicted times are trivial for the other two downstream
intersections, i. e. 4th Ave and 5th Ave. The performance of this example trip is illustrated
in
Final Report for TO 6407
14
Table 3- 4.
To further evaluate the benefit obtained from the proposed methodology, a hypothetical
trip under original signal timings was constructed and its performance was compared
with the scenario under the proposed signal timings. More than 16 seconds can be saved
for this example trip at 3rd Ave ( see Table 3- 6). More specifically,
a) If no priority request is available, the trolley will face the second half of red at 3rd Ave.
However, this trolley passed through all three signals without any stop due to the
successful execution of signal priority requests;
b) A dedicated ‘ green band’ ( not too wide) along the trolley’s direction guaranteed such
non- stop movement;
c) At the same time, a wide ‘ green band’ in the other direction made sure that the
priority execution would not affect the trolleys’ movements from the opposite
direction.
Final Report for TO 6407
15
Table 3- 4 Performance of an example trip from Civic Center to 5th Ave
3rd Ave 4th Ave 5th Ave
Pred. Leave Time 07: 09: 43 07: 09: 59 07: 10: 08
Act. Leave Time 07: 09: 46 07: 09: 57 07: 10: 09
Pred. Error ( sec) - 3 2 - 1
Block Travel Time ( sec) 11 12
Table 3- 6 Original and proposed timings for the example trip
3rd Ave 4th Ave 5th Ave
FO 2 * FO 4 ** FO 2 FO 4 FO 2 FO 4
Before
Timings 0 34 0 34 0 34
After
Timings 23 52 0 32 0 37
Expected
Delay ( sec) >= 16 0 0
However, not all trips with priority request execution gain such satisfactory results and
not all results under proposed signal timings are consistently better than those in the
original scenario. The summary of all trips is presented below.
As is shown in
Final Report for TO 6407
16
Table 3- 7, TSP successfully reduced by about 10% the number of stops along Section I
( between the American Plaza Station and the Civic Center Station). The standard
deviations are comparable for the same section. Insignificant benefits can be obtained
with applications of ATSP methodology for Section II, while minor negative impacts on
the number of stops along Section III are present. In the opposite direction, the results are
similar. TSP reduced by about 10% along Section I and reduced another 15% along
Section II. Along Section III, TSP was unable to significantly benefit trolley operation.
Final Report for TO 6407
17
Table 3- 7 Summary of number of stops at signals
Stage Section I Section II Section III
Inbound trips Mean STD Mean STD Mean STD
1 ( No TSP) 1.78 0.82 0.83 0.63 1.00 0.79
2 ( with TSP) 1.61 0.81 0.83 0.64 1.33 0.96
Outbound trips Mean STD Mean STD Mean STD
1 ( No TSP) 1.29 0.86 0.95 0.61 2.38 0.78
2 ( with TSP) 1.19 0.88 0.79 0.61 2.38 0.79
The impact of TSP on trolley travel time is similar with that on the number of stops as shown in
Final Report for TO 6407
18
Table 3- 7. The benefits were insignificant for most of trips due to some external and
internal issues, among which, the inaccurate prediction of train departure time is one of
the most important. The further detailed analysis is in the following section.
At stage 2 with TSP, some of the priority requests may be blocked due to an earlier
priority request execution for the other trolleys. To quantify the percentage of priority
requests not blocked, a priority request ‘ non- blockage’ rate, δ, is defined at a section
level ( a section is defined as the segment between two consecutive stations).
For Section i, ‘ non- blockage’ rate of priority requests is
δi = # of trips with executed priority requests / # of trips
Final Report for TO 6407
19
Table 3- 9 presents the results of this rate for different trip directions ( outbound and
inbound). As is shown in the figure, the ATSP request ‘ non- blockage’ rate is greater than
0.9 in most cases, which means that over 90% priority requests can be executed within
the scale of the field operation test. With the larger scale deployment, a higher request
blockage rate may be expected. However, based on schedule adherence, if only those late
trolleys ( around 10% of overall trips) sent out the priority request, the ATSP request
blockage rate should also fall into an acceptable range.
Final Report for TO 6407
20
Table 3- 9 Request non- blockage rates
Trolley # 1 Trolley # 2 Trolley # 6 Trolley # 8
Section
OB IB OB IB OB IB OB IB
Sec. I 0.92 1.00 0.91 0.96 0.88 1.00 0.95 1.00
Sec. II 1.00 1.00 0.98 1.00 1.00 1.00 0.95 1.00
Sec. III 0.96 0.81 0.98 0.92 1.00 0.92 1.00 0.92
A priority request consists of a set of force- off points for the intersections between
downstream and upstream trolley stations. With the changes of force- off points, the starts,
ends, and durations of signal phases would all change. When under priority, signal phase
2 serving the trolley movement direction would be relocated to cover the trolley’s arrival
time and elongated to cover the deviations of the trolley’s arrival.
Table 3- 11 summarizes the impacts on signal cycles for the “ after” scenario. Among the
total of 1234 cycles, the number of effective calls at different intersections varies due to
the blockage effects and correction calls with updates on arrival predictions. Given the
current FOT setup with limited trains under priority, the percentage of impacted cycles is
low, which is around 7%. It is noted that the transit priority should be conditioned, e. g. by
schedule adherence, if traffic operators or city engineers would like to set an upper bound
percentage of impacted cycles in order to limit the total impacts on traffic signal
coordination.
Table 3- 11 Summary of impacts on signal cycles
Total Number of Cycles Effective Calls Percentage of impacted cycles
India St 1234 105 8.5%
Front St 1234 96 7.8%
1st Ave 1234 100 8.1%
2nd Ave 1234 96 7.8%
3rd Ave 1234 86 7.0%
4th Ave 1234 92 7.5%
5th Ave 1234 91 7.4%
6th Ave 1234 89 7.2%
7th Ave 1234 87 7.0%
8th Ave 1234 86 7.0%
9th Ave 1234 94 7.6%
10th Ave 1234 87 7.0%
11th Ave 1234 85 6.9%
Final Report for TO 6407
21
Table 3- 13 summarizes the changes in durations and force- off points in phase 2 that serve
the trolleys’ direction. For all the intersections, the average changes on phase 2 durations
are positive because the proposed ATSP system tries to minimize the expected trolley
delay given that the trolleys’ arrivals are random. Both of the changes in durations and in
percentages varied among intersections. The maximum average change is about 12
seconds at 9th Ave, while the largest percentage of change is about 35% at 8th Ave, as
shown in Figure 3- 3. Although the changes within the two priority cycles are around 7
seconds and 20%, which is not low, the changes over a whole day considering the total
number impacted cycles are still very low, which are around 2%, e. g. 2.5% at India Street
as shown in Figure 3- 4. The results for other intersections can be found in the Appendix.
Table 3- 13 Summary of changes on phase 2 ( trolley phase)
Phase 2 Duration Phase 2 Force- Off ( FO)
Original
duration
( sec)
Average
change by
priority
( sec)
Change for
priority
cycles
(%)
Change
over a day
(%)
Original
FO
Average
change by
priority
( sec)
Change
over a day
( sec)
India St 37 10.87 29.4% 2.50% 0 17.1 1.45
Front St 31 7.82 25.2% 1.96% 0 10.6 0.83
1st Ave 30 4.57 15.2% 1.24% 0 12.0 0.97
2nd Ave 31 8.23 26.5% 2.06% 0 13.2 1.02
3rd Ave 32 5.32 16.6% 1.16% 0 10.4 0.72
4th Ave 32 4.70 14.7% 1.10% 0 14.5 1.08
5th Ave 32 5.99 18.7% 1.38% 0 17.9 1.32
6th Ave 32 6.08 19.0% 1.37% 0 17.3 1.25
7th Ave 32 6.32 19.7% 1.39% 0 11.4 0.80
8th Ave 24 8.44 35.2% 2.45% 0 9.7 0.68
9th Ave 37 11.97 32.3% 2.46% 0 14.3 1.09
10th Ave 33 5.82 17.6% 1.24% 0 8.8 0.62
11th Ave 28 6.90 24.6% 1.70% 0 15.3 1.05
Final Report for TO 6407
22
Figure 3- 3 Changes on phase 2 at 8th Ave
Figure 3- 4 Changes on phase 2 at India Street
Final Report for TO 6407
23
The changes on force- off points of phase 2 reflect how much priority requests shift from
the original timings. As illustrated by Table 3- 13, the average changes by priority are
approximately14 seconds compared with the original force- off point for phase 2 at the
beginning of the local clock. The changes during the priority cycles are significant in
order to substantively reduce trolleys’ delay. However, the changes over a day
considering the total number of impacted signal cycles are negligible and only 1.7%.
3.3.1.2 Impacts on Traffic Operation
The major concern for transit signal priority is the impact on general traffic. Traffic
engineers from the City of San Diego worry about the incurred delay and number of stops
for general traffic by providing transit signal priority at busy C Street. Table 3- 15
summaries the changes on phase 4 for general traffic. Among all the intersections, the
average change in duration of phase 4 is 4.9 seconds. The largest average change is 7.1
seconds at India Street, as shown in Figure 3- 5. Although the average change in duration
of phase 4 is around 20% that is significant within the two priority cycles, the impacts
over a whole day considering the number of impacted cycles per day is only 1.3%, which
is negligible. The distribution of signal changes on phase 4 at other intersections can be
found in the Appendix.
Table 3- 15 also presents the changes in phase 4 force- off points, which are an indicator of
how much the priority requests shift the signal timings from the original signal
coordination. Across all the testing intersections, the average change of phase 4 force- off
points is 6.7 seconds. The maximum average change is 9.35 seconds at India Street.
Although the change is significant over the two priority cycles, the average change over a
whole day is only 0.5 second, which is very small and negligible. According to the testing
log files, trolleys with extensive long dwell time generated multiple requests. With more
strict constraints on the number of requests for one trolley trip, the impact on other traffic
can be further reduced.
Table 3- 15 Summary of changes on phase 4 ( general traffic)
Phase 4 Duration Phase 4 Force- Off ( FO)
Original
duration
( sec)
Average
change by
priority
( sec)
Change for
priority
cycles
(%)
Change
over a
day
(%)
Original
FO
Average
change by
priority
( sec)
Change
over a
day
( sec)
India St 18 7.10 39.5% 3.36% 34 9.35 0.80
Front St 30 5.01 16.7% 1.30% 35 7.18 0.56
1st Ave 31 3.27 10.6% 0.86% 36 5.45 0.44
2nd Ave 30 5.54 18.5% 1.44% 35 6.50 0.51
3rd Ave 29 3.16 10.9% 0.76% 34 4.78 0.33
4th Ave 29 3.10 10.7% 0.80% 34 5.62 0.42
Final Report for TO 6407
24
5th Ave 29 4.76 16.4% 1.21% 34 6.82 0.50
6th Ave 29 4.67 16.1% 1.16% 34 6.52 0.47
7th Ave 29 4.31 14.9% 1.05% 34 9.16 0.65
8th Ave 37 6.31 17.1% 1.19% 42 6.08 0.42
9th Ave 24 7.03 29.3% 2.23% 29 8.37 0.64
10th Ave 28 3.97 14.2% 1.00% 33 5.10 0.36
11th Ave 33 5.22 15.8% 1.09% 34 6.03 0.42
Figure 3- 5 Changes on phase 4 at India Street
3.3.2 Prediction Analysis
The current algorithm was initially built for ‘ short- term’ ( i. e. nearest signal) prediction. It
aimed for applications with the capability of making instant changes on force- off points.
A dynamic predicted arrival time to the nearest signal is calculated by combining both
current trolley speed and historic trolley travel time. The predicted arrival time to the
prioritized signal is the sum of the dynamic predicted arrival time to the nearest signal,
the average ‘ historic’ non- stop travel time between the nearest signal and the prioritized
signal, and the dwell time at stations in between. It is noted that the trolley is assumed to
travel continuously between consecutive signals when no stations are in place.
Final Report for TO 6407
25
3.3.2.1 GPS Reception
In the FOT, the cell phone- based AVL systems ( Figure 1- 2) were installed on selected
trolley trains. Such systems failed to function as expected as shown in Figure 3- 6 where
both of the two outbound Orange Line trips deviate substantially from the tracks,
particularly at the two corners of C Street, where America Plaza Station and City College
Station are located. Figure 3- 7 illustrates two Blue Line trajectories with similar and
consistent reception issues.
Figure 3- 6 GPS trajectories for two Orange Line trips
Figure 3- 7 GPS trajectories for two Blue Line trips
The bad receptions are mainly due to two reasons. First, the cell phone- based AVL
system does not have an external antenna for the GPS receiver and limits the capabilities
Final Report for TO 6407
26
of the device to obtain good satellite information. Second, the GPS receivers at the testing
site in downtown San Diego experience the so- called “ urban canyon” effect. An urban
canyon is an artifact of an urban environment similar to a natural canyon. It is manifested
by streets cutting through dense blocks of structures, especially skyscrapers. Urban
canyons have an impact on radio reception, particularly reception of GPS signals.
Moreover, the tracks around America Plaza Station have a glass roof, which also
significantly impacts GPS reception.
3.3.2.2 Motion Prediction
Figure 3- 8 shows a typical trolley trip ( inbound trip for Figure 1). As shown in the figure,
the trolley usually stops for a long time between the predicted starting point and the 1st
test signal. This observation is quite different from the assumptions of prediction. As
shown in the figure, the first test signal for the inbound trip is C St at 11th Ave ( signal
C13). The inbound trolley first stopped at Park & Market Station for about 45 seconds
and then stopped at first station ( City College) for 87 seconds. Such discrepancies
between reality and assumption create large prediction errors for the predicted arrival
times at the first test signal and start a “ chain reaction” along downstream signals. For
example, the trolley also stopped before signal C12 ( C St at 10th Ave), which is the
downstream signal of signal C13 due to a large prediction error of arrival time at signal
C12.
Figure 3- 8 A typical trolley trajectory
Final Report for TO 6407
27
Figure 3- 9 and Figure 3- 10 show the histogram of prediction error. When the assumption
of prediction is met, i. e. no stop in between stations, the prediction is very accurate.
Otherwise large prediction errors dominate.
Figure 3- 9 Distribution of prediction errors without stopping time at stations
Figure 3- 10 Distribution of prediction errors with stopping time at stations
Final Report for TO 6407
28
3.3.2.3 Dwelling Prediction
The prediction of train dwell time is very challenging. Because all the trolley stations in
downtown San Diego are near- side stations, the total trolley dwell time is the sum of
passenger loading time, door open/ close time, and signal waiting time. As discussed in an
earlier report under Task Order 5407, the passenger loading time is highly random
because of unpredictable passenger arrivals and passenger activities. In spite of fixed-timing
control, the signal waiting time at stations is also random because train arrivals
plus passenger loading time is random.
Some observations can be made together with conclusions from the data analysis. The
distribution of dwell times at trolley stations do not have time- of- day patterns, as shown
in Figure 3- 11 and Figure 3- 12 for America Plaza Station and 5th Avenue Station,
respectively. Distribution of dwell time at some stations shows “ dual- layer” phenomena,
as shown in Figure 3- 12. The average time difference of the two “ layers” is around 70
seconds, which is exactly a full signal cycle. Such phenomenon shows that the trolley
arrival times at the 5th Avenue Station normally situate at a similar location on the local
signal control clock for the downstream intersection. The departure time would be either
the next start of green or the following green if the trolley cannot finish loading
passengers by the beginning of green. Operators actually follow the rule of departing
stations only at a fresh green. It is noted that such phenomenon normally happens when
TSP is not activated.
Figure 3- 11 Distribution of dwelling time at America Plaza Station
Final Report for TO 6407
29
Figure 3- 12 Distribution of dwelling time at 5th Ave Station
3.3.2.4 Analysis of Prediction Errors
The prediction errors are mainly contributed by four factors: passengers’ activities,
operator behavior, equipment accuracy, and traffic signal operations. Although detailed
GPS trajectory data and traffic operation data are collected, the prediction error cannot be
directly measured because the exact time when a trolley is ready to depart from a station
is unknown. Here we analyzed the prediction errors by following four randomly selected
sample trolley trajectories.
The first chosen trip was an outbound train entering the testing site at 6: 59: 54 on October
16. The first predicted time- to- arrival ( TTA) to India Street started at 78 seconds. From
7: 01: 09 to 7: 06: 54 for 345 seconds, the train’s GPS location almost did not move at all. It
is totally different from the historical dwell time at America Plaza Station. The predicted
TTA stayed at about 18 seconds from 7: 01: 09 to 7: 04: 27 and jumped to 34 seconds at
7: 07: 20. The reason for the failure prediction is the extensive long dwell time and
possible bad GPS reception under the glass roof at America Plaza Station.
The second selected trip was an inbound trip started at 9: 30: 14 on October 16, 2009. The
first predicted TTA to 11th Ave started at 69.5 seconds. The train didn't stop at signals
before 11th Ave. and departed at 11th Ave station at 8: 32: 26. Given the historical dwell
time of 31 second at City College and 21.6 seconds at Market Street, the prediction error
is only 10 seconds and within 10%. The train left 11th Ave at the beginning of the green
Final Report for TO 6407
30
cycle. Because of the predicted 10 seconds early, the train stopped for about 5 seconds at
7th Ave and went through all other intersections without any stops.
The third picked trip was an inbound trip started at 10: 23: 06 on October 16, 2009. The
predicted TTA to 11th Ave started at 67.5 seconds. The train didn’t stop at the signal
before 11th Ave but stopped at Market Street station from 10: 23: 26 to 10: 24: 21 for 55
seconds, which is much longer than the historical time of 21.6 seconds. The train
departed 11th Ave at 10: 26: 37 at the beginning of the green cycle. Given the sum of two
historical dwell times of 52.6 seconds, the prediction error can be 91 seconds. But the
operator might stop the train to wait for a fresh green rather than take advantage of TSP.
The waiting might take up to 70 seconds. Not surprisingly, TSP did not benefit the trolley
operation for this trip. However, the impact on general traffic still existed.
The last picked trip was an outbound trip started at 14: 10: 51 on October 16, 2009. The
predicted TTA to India Street started at 95.0 seconds. The trolley stopped from 14: 11: 39
to 14: 13: 05 for 86 seconds and then moved very slowly to reach India Street at 14: 17: 09,
which is 378 seconds after the train enters the boundary of the testing site. Given the
historical dwell time of 15.6 seconds and 21.3 seconds at Santa Fe and America Plaza,
respectively, the predicted departure time is totally wrong. It is partially because of the
extensively long dwell time at Santa Fe and also at America Plaza. Because the GPS was
slowly moving, it is hard to tell when the train stopped at America Plaza. GPS was the
issue under the roof at America Plaza station.
It is noted that the success of train arrival prediction would normally lead to a successful
TSP implementation, as illustrated by the second picked trip. However, many cases have
significant issues in predicting the departure time at those near- side stations. Some
observations can be summarized here:
• Extensive long dwell time at stations ( our observations underestimate dwell time
in most cases)
• GPS reception issue at America Plaza due to the glass roof.
• Train might stop at crossings before arriving at America Plaza.
• Train might stop at signals before arriving at City College.
Final Report for TO 6407
31
4. Recommendations and Next Step
The proposed system design can be easily applied to other light- rail transit ( LRT)
systems, which do not have the preempted right- of- way at grade crossings and
intersections. It is not necessary to have fixed- timing control at the signalized
intersections. The proposed concept and system can be easily adapted to actuated or
semi- actuated control systems. Moreover, the results and lessons learned from both of the
laboratory test and the preliminary field operational test could help in designing and
calibrating such transit signal priority systems for other similar LRT systems.
The preliminary FOT has been completed. According to the data analysis, there are still
many issues before a large deployment of the proposed ATSP system may be made. This
section summarizes the issues and recommendations in order to further improve the
system towards the next step. With all the system improvements and testing, a final FOT
will be performed.
4.1 Signal Transition
As described in a previous report under Task Order 5407, the proposed adaptive signal
control strategy ( Scheme I) is used for those late trips, which account for about 10% of
total trips. If the system capacity is required to be increased, say, there are 80% of trips,
which are late or require adaptive priority, then the strategy shown in the previous section
will fail to work. The major restriction results from the logic of signal controllers, in
particular, signal transition logic.
In many cases, an additional cycle is required for signal controllers to complete the
transition from one set of signal timings to another. Therefore, if the frequency of a
priority request increases, then such a transition period becomes longer, which will have
more negative impacts on the whole traffic system, e. g. unrequested trolleys and cross-street
traffic.
In addition, due to the signal transition logic, the solution of the proposed adaptive signal
control model may not be implementable in the field. For implementation, the signal
timings in the current cycles may highly relate to the signal timings in the previous cycle.
There are at least two remedies to take into account the signal transition logic:
• Set up another model to obtain the signal timings for the transition cycle, such
that the signal timings from the proposed adaptive signal control strategy are
guaranteed to be implemented in the cycle after the transition;
• Put more constraints on the adaptive signal control model mentioned above, such
that the signal timings from the modified adaptive signal control strategy are to be
implemented in the cycle right after the one with base- line timings.
Final Report for TO 6407
32
4.2 Signal Progression
The original signal progression design also impacts the performance and design of the
TSP system. The better the original progression design requires little timing change to
redesign the progression for approaching trolleys when given real- time trolley movement
information. According to the field data, some segments actually suffer from the existing
signal progression design. As shown in Figure 4- 1 and Figure 4- 2, many trajectories have
to stop at intersections between stations due to inappropriate progression design. The
progression results for other segments are included in the Appendix. Therefore, it is also
important to redesign the signal progression before the large- scale implementation of the
priority system.
Figure 4- 1 Outbound trajectories between America Plaza and Civic Center ( No TSP)
Final Report for TO 6407
33
Figure 4- 2 Inbound trajectories between Civic Center and America Plaza ( No TSP)
4.3 Dwell Time Prediction
The proposed adaptive signal control strategy takes the trolley’s predicted arrival time as
one of the inputs. The effectiveness of the proposed strategy largely depends on the
accuracy of the arrival time prediction. Unfortunately, few studies have been conducted
on the prediction of station dwell time because there are so many uncertainties which
make it impossible to obtain an accurate prediction. For example, if a handicapped person
needs to board the trolley, the dwell time may be much longer ( e. g. 2 or 3 more minutes)
than usual. In addition, the downstream signal status also contributes to the dwell time of
nearside stations.
However, a simple linear regression model based on limited observed field data is applied
to predict the dwell time at each station. More data from the field are required to obtain
more knowledge on the dwell time. Furthermore, the prediction of dwell time statistics,
e. g. 95% percentile, is more tractable and pragmatic than the prediction of exact dwell
time or the mean of dwell time. Due to the interaction between the trolley dwell time and
the downstream signal status, it is more appropriate to optimize the signal timings by
integrating them with dwell time prediction.
4.4 Trolley’s Arrival Time Prediction at Station
As presented in the previous section of this report, the TSP benefits on trolley operation,
particularly on trolley travel time, are not as expected. Under the existing ATSP strategy,
Final Report for TO 6407
34
the priority algorithm takes predicted trolley departure times at stations, the current signal
status at downstream intersections and signal timing constraints as inputs to design a
desired trolley green band. In order to have local controllers to execute the trolley green
band timing, the timing decision needs to be made 2 minutes ahead of start time of the
designed green band. Due to the variations of a trolley’s intersection delay and dwell
times at a station, a trolley often misses the designed green band.
The predicted trolley departure time at a station consists of two components, the
predicted arrival time at a station and the predicted dwell time at a station. The trolley’s
intersection delay was not considered when providing the first component - the predicted
arrival time at a station and a predetermined constant was used as the second component
– the predicted dwell time at a station. The inability of the prediction algorithm in dealing
with variations of trolley intersection delay and dwell time is the major cause of a trolley
missing the designed green band.
Real- time signal status needs to be incorporated into the prediction algorithm to gain the
ability of estimating a trolley’s intersection delay. As a trolley does not share the road
with general motor vehicles, it is reasonable to believe incorporating signal status will
achieve a more accurate prediction.
4.5 Integrating Priority Decision with Prediction
Dwell time at a station has variability. Along the testing segment, the width of priority
green band is usually less than 20 seconds, with a 60 second cycle length. It is not
practical to precisely predict the dwell time so that it can meet the requirement of the
priority algorithm. A more practical approach is to predict the probability of “ can
departure”, conditional on the time window, the time from the predicted arrival time at
station to the start time of the green band at the downstream intersection, and to integrate
the priority decision with the prediction. This approach does not require to precisely
predict a trolley’s dwell time at a station but rather to influence a trolley’s stop time at a
station ( combination of dwelling time and green signal waiting time) to minimize the
travel time between stations.
To illustrate how influencing stop times at a station affects the probability of a trolley
missing the green band, Figure 4- 3 below shows the comparison of trolley stop time at
City College station for westbound trips ( left- side plot) and eastbound trips ( right- side
plot), under the without TSP scenario. There are two groups of blue bars in both plots.
The group on the left side corresponds to trips upon departure at the coming green light
while the group on the right side corresponds to trips that missed the coming green and
waited for one more signal cycle.
Final Report for TO 6407
35
Figure 4- 3 Trolley stop time at City College Station ( No TSP)
Compared with westbound trips, eastbound trips have slightly longer stop times upon
departure at the coming green and with much less chance of missing the coming green
( 10 percent vs. 40 percent). The difference in stop time distributions for west- and east-bound
directions ties with the underlining signal timing. Figure 4- 4 shows west- and east-green
bands along the test segment. In the plot, westbound trips travel downward. At City
College Station - the solid black line in the middle, the offset between the upstream green
band and downstream green band on westbound is obviously smaller than that on
eastbound. Westbound trips have less chance to finish passenger loading/ alighting
activities within the offset and therefore have a higher chance of missing the coming
green band.
Figure 4- 4 Green bands along testing segment ( No TSP)
There is a tradeoff between signal waiting time at a station and travel time to the
downstream station, i. e., the time from arrival at a station to arrival at its downstream
station. The longer the waiting time is, the higher the chance that a trolley could hit the
designed green band but could result in longer travel time to the downstream station. The
Final Report for TO 6407
36
integrated approach dynamically makes the tradeoff decision to minimize the total travel
time.
4.6 Automatic Vehicle Location ( AVL) System
The cellphone- based cost- effective AVL system in the preliminary FOT is not successful
due to the serious “ urban canyon” effect. A better solution should be proposed and/ or
tested. At the beginning of the project, another type of AVL system based on a GPRS
modem and GPS receiver with external antenna was tested. As shown in Figure 4- 5, the
trolley trajectories are more stable and closer to the geometry street map when compared
with the results from the cellphone- based system. Such a system with an external GPS
antenna can be a potential solution for future testing. However, more testing is still
needed before the next FOT, particularly at the area around America Plaza Station where
the existing system performed the worst.
Figure 4- 5 GPS receptions with GPS external antenna
Final Report for TO 6407
37
Appendix
Table 5- 1 Detailed trip samples for Stage 1 without TSP
Trolley # 1 Trolley # 2 Trolley # 6 Trolley # 8 Summary
Date
OB IB OB IB OB IB OB IB OB IB
30th 0 0 0 0 1 2 6 8 7 10
31st 0 0 0 0 0 0 0 0 0 0
1st 0 0 0 0 0 0 6 7 6 7
2nd 0 0 0 0 3 4 6 7 9 11
3rd 1 1 0 0 2 3 2 4 5 8
4th 1 2 3 2 3 3 4 5 11 12
5th 1 2 0 1 3 4 7 8 11 15
6th 1 2 6 7 3 4 9 8 19 21
7th 0 0 0 0 0 0 5 6 5 6
8th 0 0 5 6 0 0 2 2 7 8
Sum. 4 7 14 16 15 20 47 55 81 98
Final Report for TO 6407
38
Figure 5- 1 Changes on phase 2 ( trolley) at Front Street
Figure 5- 2 Changes on phase 2 ( trolley) at 5th Street
Final Report for TO 6407
39
Figure 5- 3 Changes on phase 2 ( trolley) at 6th Street
Figure 5- 4 Changes on phase 2 ( trolley) at 7th Street
Final Report for TO 6407
40
Figure 5- 5 Changes on phase 2 ( trolley) at 8th Street
Figure 5- 6 Changes on phase 2 ( trolley) at 10th Street
Final Report for TO 6407
41
Figure 5- 7 Changes on phase 2 ( trolley) at 11th Street
Figure 5- 8 Changes on phase 4 ( trolley) at Front Street
Final Report for TO 6407
42
Figure 5- 9 Changes on phase 4 ( trolley) at 5th Street
Figure 5- 10 Changes on phase 4 ( trolley) at 6th Street
Final Report for TO 6407
43
Figure 5- 11 Changes on phase 4 ( trolley) at 7th Street
Figure 5- 12 Changes on phase 4 ( trolley) at 8th Street
Final Report for TO 6407
44
Figure 5- 13 Changes on phase 4 ( trolley) at 10th Street
Figure 5- 14 Changes on phase 4 ( trolley) at 11th Street
Final Report for TO 6407
45
Figure 5- 15 Outbound trajectories between Civic Center and 5th Ave ( No TSP)
Figure 5- 16 Outbound trajectories between 5th Ave and City College ( No TSP)
Final Report for TO 6407
46
Figure 5- 17 Inbound trajectories between City College and 5th Ave ( No TSP)
Figure 5- 18 Inbound trajectories between 5th Ave and Civic Center ( No TSP)