1. ABSTRACT
Title of Thesis: EVALUATING EFFICIENCY OF DYNAMIC
BUS ROUTING IN BALTIMORE CITY
Grafton Henry Carter Ray IV, Masters of
Professional Science in Geographic Information
Science, 2015
Thesis Directed By: Professor Dr. Eunjung Lim, Department of
Geographical Sciences
Traffic congestion is a major cause of delays in bus transit scheduling.
Alleviating/mitigating these delays is a primary goal of transportation agencies. This
research paper attempts to discover whether or not dynamic bus routing can conserve
bus schedule routing time windows in Baltimore City throughout various levels of
traffic congestion. The research is modeled on a specific northbound bus route
traversing Baltimore City. The research employs vehicle routing analysis using ESRI
ArcGIS network analyst to determine the feasibility as well as the efficiency of
dynamic bus routing using bus route and scheduling data obtained through the
Maryland Transit Administration (MTA). The results show that during peak periods
of traffic volume, dynamic bus routing is able to conserve bus schedule routing time
windows in Baltimore City. Dynamic bus routing shows to be a viable solution to
improving public bus transportation in high density, highly-congested urban areas.
2. DETERMINING EFFICIENCY OF DYNAMIC BUS ROUTING IN BALTIMORE
CITY
by
Grafton Henry Carter Ray IV
Thesis submitted to the Faculty of the Graduate School of the
University of Maryland, College Park, in partial fulfillment
of the requirements for the degree of
Masters of Professional
Science in Geographic
Information Science
2015
Advisory Committee:
Professor Dr. Eunjung Lim, Chair
4. ii
Dedication
This thesis is dedicated to my wife NATALIE who has lovingly supported me
throughout all of my academic endeavors.
To my son PARKER, for through the pursuit of knowledge we all strive to make the
world a better place for future generations.
To Dr. Lim and all of the professors, researchers, and staff at the University of
Maryland Department of Geographical Sciences who have mentored and guided me
throughout my academic career.
5. iii
Table of Contents
Dedication..................................................................................................................... ii
Table of Contents.........................................................................................................iii
List of Tables ............................................................................................................... iv
List of Figures............................................................................................................... v
List of Equations.......................................................................................................... vi
Chapter 1: Introduction................................................................................................. 1
Background............................................................................................................... 1
The Case for Bus Transit .......................................................................................... 1
The Problem with Bus Transit.................................................................................. 2
Dynamic Bus Routing............................................................................................... 4
Objective................................................................................................................... 4
Chapter 2: Data ............................................................................................................. 6
Study Area ................................................................................................................ 6
Streets Layer ............................................................................................................. 6
Bus Stops .................................................................................................................. 8
Bus Lines .................................................................................................................. 8
Baltimore City Traffic Data...................................................................................... 9
Bus Time Schedule ................................................................................................. 10
Chapter 3: Methodology ............................................................................................. 12
Data Processing and Selection................................................................................ 12
Building the Network Dataset................................................................................. 13
Vehicle Routing Analysis ....................................................................................... 16
Chapter 4: Results....................................................................................................... 20
Control Routing Analysis ....................................................................................... 20
Routing Analysis Using K-Factor........................................................................... 21
Chapter 5: Conclusion/Discussion.............................................................................. 23
Conclusion .............................................................................................................. 24
Discussion............................................................................................................... 24
Appendix..................................................................................................................... 26
Bibliography ............................................................................................................... 34
6. iv
List of Tables
Table 1 – Street Classification...................................................................................... 7
Table 2 - Ft McHenry - Saini Hospital Northbound Bus Schedule............................ 11
Table 3 - Drive Time Violations without Traffic........................................................ 20
Table 4 - Drive Time Violations with Traffic............................................................. 22
Table 5 - Wait Drive Time.......................................................................................... 22
7. v
List of Figures
Figure 1 - Baltimore City Study Area........................................................................... 6
Figure 2 - Baltimore City Streets Data Layer............................................................... 7
Figure 3 - Bus Stops Data Layer................................................................................... 8
Figure 4 - Bus Route Data Layer.................................................................................. 9
Figure 5 - K-Factor ....................................................................................................... 9
Figure 6 - Removed Bus Stop..................................................................................... 12
Figure 7 - Global Turn Delay Evaluator..................................................................... 15
8. vi
List of Equations
Equation 1 - Street Hierarchy Attribute Evaluator ..................................................... 14
Equation 2 - Drive Time Attribute Evaluator ............................................................. 14
Equation 3 - K-Traffic Field Calculation.................................................................... 18
9. 1
Chapter 1: Introduction
Background
Thousands of people in urban centers around the United States rely on public
transportation every day. The most wide spread mode of transportation used is in
these systems is bus service. Buses are a relatively cheap, reliable, and easily
deployable way to move large quantities of passengers. Buses are easily incorporated
into transit systems due to their low level of start-up capital investment, low operating
cost, and the ability of routes and stops to be changed. Heavy and light rail systems
require large capital investment for construction and maintenance. Buses on the other
hand do not require any capital investment other than the buses themselves. Bus
systems travel along city streets, and therefore unlike fixed rail, are not limited in
their ability to reach any destination. This also allows for the more frequent
placement of stops along routes, as well as the ability of the routes to be changed in
the future; fixed rail can only travel along laid track. This allows bus transit to be
more versatile, cost effective, and more manageable for urban centers.
The Case for Bus Transit
Bus transit ridership has increased over the past decade. In Great Britain, bus
transit ridership rose by 8% between 1999/2000 and 2005/2006 (White, 2009). With
increasing ridership, as well as increasing populations, bus transit is becoming the
most popular mode of urban mass transit. Busses are cheaper than rail; the average
light rail line being built today cots about $100 million per mile; including the cost of
stations, park-and-ride lots, and other infrastructure but not rail cars (O’Toole, 2014).
10. 2
Since bus transit operates on existing streets, no infrastructure (other than bus depots,
and physical stop maintenance) needs to be built. This drastically lowers the cost
capital cost of bus transit systems.
Another advantage of bus transit of rail is capacity. The standard 40-foot bus
has about 40 seats and room for 20 standing; for a total of 60 passengers, at an
average cost of $300,000 to $400,000. Compare this to the cost of $4.5 million for a
light rail car, at just under 100-feet, and the bus transit cost advantage is obvious.
Bus transit can carry more people to more places, for less.
The Problem with Bus Transit
Scheduling is a large component of any transit system. Scheduling is required
for the bus transit system to be of any use to its customers. The term “customers” is
used here to describe bus transit riders. This term is chosen because riders pay a fare,
or user fee, to use the transit system. They are essentially buying a product, or paying
to use the bus transit service. Bus system operators, whether they be public agencies
or private entities, seek to attract users as to collect revenue. Therefore the operators
aim to maximize the efficiency of the bus transit system. Efficiency in this paper will
be defined as the timeliness of the transit system operation. The efficiency of the bus
transit system will be defined as the ability of a bus on a particular route to reach each
stop within a scheduled time window. A bus transit system that does not run on time,
is of no use to its customers because the purpose of using the bus system for transit is
useless if a customer does not know what time a bus will arrive at a particular stop. It
is also just as useless if the bus does not, in practice, arrive at a particular stop at the
scheduled time.
11. 3
The main problem involved with bus transit involves the same aspect that
allows the bus system to be so versatile, the street network. Since busses use the
same streets as other vehicular traffic, they are subject to the same issues and delays
that all other traffic is subject to. The advantage of fixed rail is exactly that, fixed
rails. These rails are solely used by the rail transit system, and barring any signaling
issues or track maintenance, provide a dedicated route of travel clear of congestion. It
is possible to have dedicated bus transit lanes on larger roads and highways, however
this is impractical on smaller urban residential streets. Therefore, bus transit lines are
at the mercy of traffic and road closures. Since buses usually follow a fixed route
between stops in a fixed order as to travel the shortest distance between stops, any
delay or obstruction in this route will cause the disruptions in the transit system.
These disruptions can slow the bus down, effectively causing the bus to be late to a
particular bus stop. In actual operations, stochastic bus travel times often occur,
hampering the performance of planned bus routes and schedules (Yan, et. al, 2008).
These issues are hard to predict and do not usually planned for during the route
planning stage. If a bus is delayed and does not reach a stop within the scheduled
time window, this delay propagates downstream, and subsequent stops will not be
reached on time (Yan, et. al, 2008). This causes the delay, and time window
violation, to become greater the farther downstream on the route. Time window
violations that become too great render the bus transit system unusable by the
customers.
12. 4
Dynamic Bus Routing
Dynamic Bus Routing is the concept of changing bus routes based on
stochastic factors that occur along the bus route, in an attempt to preserve the
scheduled time windows for each stop. The stochastic elements usually examined in
these problems were: customers, their demands, and travel times (A. Larsen et. al,
2002). However, for this paper, customer demands will be represented by the
scheduled time windows for each bus stop, with the bus stops representing the
customers themselves. This route change would include using different streets to
reach a bus stop. With the ability for a bus traveling along a route to avoid time
consuming hazards such as road closures, construction, or even traffic, would allow a
bus to reach its next stop without violating its scheduled time window. There is a
limit to how much time would be needed to correct or avoid a stochastic delay. For
instance, if a vehicle collision were to occur a few blocks in front of a bus, the driver
would have no way of knowing, and thus would not be able to alter the route in time.
However, if the dynamic bus route could be calculated at the beginning of a route,
then drivers would have ample time to adjust their route accordingly. This paper will
attempt to evaluate the efficiency of dynamic bus routing.
Objective
The objective of this paper will be to evaluate the efficiency of dynamic bus
routing in the City of Baltimore, Maryland. The paper will utilize vehicle routing
analysis to determine if dynamic bus routing can preserve the time window
constraints of bus routes regarding arrival times at particular bus stops. The factors
effecting the vehicle routing analysis will consist of time constraints imposed on
13. 5
street segments based on distance and peak-hour traffic volume. The paper will
attempt to determine the feasibility as well as the efficiency of dynamic bus routing
considering added time constraints imposed due to peak traffic volume. The criteria
for determining feasibility will consist of the ability of the vehicle routing analysis to
reach all the stops in order after dynamic routing. The criteria for determining
efficiency will consist on the ability of the vehicle routing analysis to reach all the
stops in order and within specified time window containing the scheduled bus arrival
time.
14. 6
Chapter 2: Data
Study Area
The study area for this project will consists of Baltimore City. Specifically
the more densely populated downtown and surrounding area. This is due to a large
concentration of bus stops and bus routes in the area. This area of the city has denser
traffic and a higher number of streets and intersections. This will ensure a variety of
options for dynamic routing of busses.
Baltimore City is the largest incorporated
city in the state of Maryland. With a
population of about 620,000 and covering
92 square miles; it is the largest urban area
in Maryland. The city has a variety of bus
lines crisscrossing the city and a myriad of
bus stops. This makes the city ideal for
testing the efficiency of dynamic bus
scheduling.
Streets Layer
The Baltimore streets data layer consists of all major and minor streets,
highways, and interstates in the city of Baltimore. The streets data layer was obtained
from the Maryland Transportation Administration (MTA) GIS Data Collection. The
streets data layer consists of a GIS polyline shapefile representing the streets of
Baltimore City. The data layer is in the NAD 83 North American Datum geographic
Figure 1 - Baltimore City Study Area
15. 7
projection. The data layer contains attributes describing the streets. The streets data
layer is used to build the streets network dataset used in the vehicle routing analysis.
Only a few of these attributes are
used in building the model. The first
attribute used is a street class attribute that
describes a hierarchy of road classes. The
hierarchy consists a numbered classification
system in which integer values represent
classes of streets. The attribute will be used
to describe the street hierarchy in the
network dataset. This will be used to
determine which streets the routing analysis
will prioritize when dynamically routing the
bus routes. The second attributed used to model the network dataset is a to-and-from
line segment value. This value represents
the elevation value at the end of each
segment. An elevation value of “1”
indicates that the end of a street segment is
elevated above other segments that it
crosses. An elevation value of “0” indicates
that the end of a street segment is coincident
Figure 2 - Baltimore City Streets Data
Layer
Code Description
1 Rural Interstate
2 Rural Other Principal
6 Rural Minor Arterial
11 Urban Interstate
12 Urban OPA Freeway/Expressway
14 Urban Other Principal Arterial
16 Urban Minor Arterial
17 Urban Collector
19 Urban Local
Table 1 – Street Classification
16. 8
with other segments that it crosses. The final attribute is a shape length attribute.
This attribute represents the length of each street segment in feet. Feet is used as the
distance unit because it is the unit used in the projection of all the datasets.
Bus Stops
The bus stops data later consist of all of the MTA bus stops in the state of
Maryland. The data layer was obtained from the Maryland iMap GIS Data Portal.
The bus stop layer consists of a GIS point
shapefile layer representing 5946 MTA bus
stops in Maryland. The bus stop data layer
is in the WGS 1984 North American Datum
geographic projection. The bus stop data
layer will be used to model the bus stops in
the vehicle routing analysis. The only
attributes of the data layer that are used in
the model are the stop locations and the stop
names. The data was selected to include
only the Fort McHenry to Saini Hospital
northbound route.
Bus Lines
The bus line data layer consists of all MTA bus routes in the state of
Maryland. The data layer was obtained from the Maryland iMap Data Portal. The
bus line data layer consists of a GIS polyline shape file layer representing bus lines in
Figure 3 - Bus Stops Data Layer
17. 9
Maryland. The bus line data layer is in the
WGS 1984 North American Datum
geographic projection. The bus line data
will be used to model the bus lines in the
vehicle routing analysis. The only attributes
of the data layer that are used in the model
are the locations of the lines and the names
of the lines. The data was selected to
include only the Fort McHenry to Sinai
Hospital northbound bus route.
Baltimore City Traffic Data
The Baltimore City traffic data layer
consist of a Baltimore city street layer
containing traffic data. The data layer was
obtained from the Maryland Transportation
Administration (MTA). The traffic data
layer consists of a polyline shapefile
representing different traffic data for each
road segment. The traffic data attribute used
for this study is the K-Factor. The K-Factor
is the proportion of Annual Average Daily
Traffic (AADT) occurring at the 30th highest hour of traffic density from the year's-
Figure 4 - Bus Route Data Layer
Figure 5 - K-Factor
18. 10
worth of data. The K-factor is defined as the proportion of average daily traffic
occurring in an hour. The higher the K-Factor, the more congested traffic becomes
on that particular road segment. The K-factor will be used to represent traffic
congestion on the road segments when evaluating dynamic bus routing.
Bus Time Schedule
The bus time schedule consist of a time table of arrival times for each bus stop
on the Fort McHenry to Sinai route. The table includes the bus arrival times at each
stop along the route for each iteration of the bus route throughout the day. The bus
does not stop at every stop on every iteration of the route. Therefore, the northbound
route beginning at 4:08 PM is selected because it services all the stops along the route
except one. It also corresponds to a high peak volume traffic during the evening rush
hour.
20. 12
Chapter 3: Methodology
Data Processing and Selection
Before any data processing can be conducted, all data layers are projected
from their original geographic projections to NAD 1983 2011 State Plane Maryland
FIPS 1900 US (feet) projected coordinate system. Changing the coordinate system
from a geographic coordinated system to a projected coordinate system ensures that
distances measured across the study area are consistent. This consistency is
necessary for setting the cost parameters when building the network dataset.
Data selection begins with the bus routes. MTA bus “Route 1” is selected and
exported to its own individual data layer. This isolates the bus route on which to run
the vehicle route analysis. The “Route 1” bus route is runs northbound from southern
Baltimore City to the northwest. This bus route is chosen because it traverses densely
congested streets in the south and
downtown area, and a more suburban
street pattern in the northwest. This
route provides a mix of street densities
on which to perform the routing
analysis (see Figure 4).
Next, the bus stop data layer is
selected to include only the bus stops
that serve “Route 1”. The resulting
selection contains ten individual bus Figure 6 - Removed Bus Stop
21. 13
stops; Fort McHenry, Charles & Cross, Fayette & Eutaw, Fulton & North,
Mondawmin Metro Station, Reisterstown & Liberty Heights, Greenspring & Cold
Spring Lane, Tamarind & Yellowwood, and Sinai Hospital. The bus route does not
stop at Mondawmin Metro Station during the 4:08 PM iteration. Therefore, this stop
is removed from the selected stops to maintain continuity among the stops and
ensuring that the bus route must stop at every stop.
The average speed of a city bus in Maryland is 11.6 miles per hour (MPH)
(Foursquare Integrated Transportation Planning & Jacobs Engineering, 2014),
therefore a speed limit field is added to the streets data layer. This speed limit will be
used in conjunction with the street length field to calculate a travel time cost for each
street segment.
Lastly, the bus schedule time table is joined to the bus stop layer. This is done
by joining the two tables based on the stop name.
Building the Network Dataset
After processing and selecting the data layers, the network data set is built.
The network dataset is built using the Baltimore City streets data layer. After
creating the network data set, attributes are assigned to the dataset. These attributes
will enable the vehicle routing analysis calculate drive times.
The fist attribute assigned to the network data set is the length attribute. This
attribute is a cost attribute which represents the total distance traveled by a bus along
the route. The units of the length attribute are in feet, and the value of the attribute is
equal to the length of a street segment. The second attribute is a street hierarchy.
This attribute represents the classification of the streets to restrict the route analysis
22. 14
from using streets that are unreasonable for the bus route. The street hierarchy
attribute evaluator is set to:
[CLASS_RTE] >= 14
Equation 1 - Street Hierarchy Attribute Evaluator
This restricts the route analysis from using any streets that are larger than an urban
arterial street. This class restriction excludes urban interstates that are limited access.
This restriction is applied in order to give the routing analysis the most available
options to dynamically re-route the bus.
The third attribute assigned to the network data set is the drive time attribute.
This attribute is also a cost attribute which represents the amount of time it takes for a
bus to traverse a road segment. The units of the drive time attribute are minuets. The
drive time attribute evaluator is set to:
(([Shape_Length] / 5280) / [SPEED]) * 60
Equation 2 - Drive Time Attribute Evaluator
The drive time attribute restricts the bus route by adding a time factor to each road
segment based on the length of the road segment, therefore it takes longer to traverse
longer road segments. The road segments travel time is cumulative, meaning that
traveling on a higher number of shorter road segments does not necessarily mean a
shorter travel time. Since the travel speed is the same along all road segments, the
distance traveled is the only factor influencing the travel time.
The final attribute built into the network dataset is a global turn attribute.
Global turn attributes are time costs imposed on intersections of road segments.
Global turns are turns that can be made at all coincident street endpoints
(intersections). This accounts for the time taken to make left and right turns, as well
23. 15
as cross over other road segments. This attribute also accounts for possible time
buses are stopped at traffic signals when crossing over other road segments. These
times are entered into the Global Turn Delay Evaluator.
No time cost is given for turns made by traveling along a road segment
connecting straight to another road segment that does not cross another road segment.
A three second time delay is given to a bus traveling from a local road to another
local road where it must cross a local road. This three second delay is used to account
for possible traffic signals that a bus may encounter. This number is relatively low
considering that a bus may not have to stop at every traffic signal when crossing an
intersecting street. A five second delay is given to a bus that must make a right turn
onto another road. This five second delay accounts for the time a bus must spend at a
stop sign as well as time waiting for traffic to clear before it is able to make a right
turn. A 15 second delay is given to a bus that must make a right turn across opposing
Figure 7 - Global Turn Delay Evaluator
24. 16
traffic onto another intersecting road segment. This 15 second delay accounts for the
time a bus spends at a traffic signal waiting for the right of way to turn left, or the
time a bus spends waiting for traffic to clear before making a left hand turn across
opposing traffic lanes. Global turn delays are designed to account for real-world time
delays encountered by buses when making turns throughout the network data set.
Therefore the more turns a route takes, the higher the time cost of the route. This is
designed to emulate the time it would take to make multiple turns, thus discouraging
the vehicle routing analysis from making “zig-zag” routes across the densely gridded
street layout of dense urban areas.
Vehicle Routing Analysis
ESRI’s ArcGIS Network Analysis extension offers multiple analysis options
to be performed on a network dataset. For this paper, the route analysis network layer
is used. Route analysis is used to solve for a route depending on the impedance
chosen to solve for. Therefore the best route is defined as the route that has the
lowest impedance, or least cost. For this paper, the impedance used to solve for a
route is the drive time cost attribute. The route analysis solves for the route that takes
the least amount of driving time.
For this analysis, two route analysis layers are used. These two layers are the
stops class and the route class. The stops class stores the network locations that are
used as the stops. For this analysis the bust stops layer will be used to populate the
stops class. The stops class stores the network locations of the bus stops, as well as
various other attributes such as start and end times and service time windows. For
this paper, both the start time and service time windows are used in conjunction with
25. 17
the bus schedule time table. The route class will be discussed later in the results
section.
Before the route analysis can be conducted the stops class must be populated.
The bus stops data layer is imported into the route analysis stops class. The stop
name attribute is populated with the individual stops name, and the time window
attribute is populated with the time that the bus arrives at the stop according to the bus
schedule time table. This instructs the route analysis that the bus must leave the first
stop at 4:08 PM, and that all subsequent stops must be reached by the time indicated
on the time table.
Second, the routing analysis settings must be configured. These settings
determine what parameters are used in evaluating the route. The impedance
parameter is set to drive time network dataset attribute. This ensures that the route
analysis will find a route that minimizes the travel time needed to reach each stop
within the time window parameter. No specific start time of the route is used. Since
the bus makes many iterations of the same route, the start time of the studied iteration
will be represented by the time constraint window of the first stop at 4:08 PM. The
bus does not need to start at any particular time before it reaches the first stop. It only
has to be at the first stop at the scheduled time. The whereabouts of the bus before
reaching the first stop along the route are beyond the scope of this study. The route
analysis layer must honor the time windows established by the bus schedule time
table. All stops must be reached by the scheduled time or a time window violation is
logged. These time window violations will be used to evaluate the effectiveness of
26. 18
the dynamically routed buses. The last parameter is set so that a bus may not make
any U-turns along the route. Finally, the routing analysis is solved.
The first routing analysis acts as a control. The first analysis does not include
the traffic data. It is used as a base line to test whether the bus arrival times can be
preserved once traffic data is added to the analysis. To add the traffic data, only one
parameter is changed. All aspects of the aforementioned methodology are identical
except the travel time network dataset attribute. The travel time attribute is altered to
account for the K-Factor traffic congestion indicator. Before the travel time attribute
can be altered, the K-Factor must be normalized over the length of the road segments.
Since each road segment is assigned a K-Factor value, traversing one segment after
another would account for each segments K-Factor, this however compounds the K-
Factors. Having the K-Factor normalized to the length of the road segment ensures
that the K-Factor is distributed proportionally over the entire road segment. This
allows for an even distribution of traffic factors over the network. A new named
[K_TRAFFIC] field is added to the streets data layer. This field is calculated as:
[K_TRAFFIC] = [K_FACTOR] / [Shape_length]
Equation 3 - K-Traffic Field Calculation
Next, the drive time network attribute is adjusted to account for the increased
traffic volume. This is done by changing the drive time attribute evaluator to add
additional time proportional to the K-Factor travel volume over the length of the
street segments. The updated drive time attribute evaluator is calculated as:
((([Shape_Length] / 5280) / [SPEED]) * 60) + [K_Traffic]
Equation 4 - Drive Time Attribute Evaluator w/K-Factor
27. 19
The routing analysis is solved, only this time with the K-Factor included in the drive
time cost attribute.
28. 20
Chapter 4: Results
Control Routing Analysis
The results of the first routing analysis without the K-Factor traffic parameter
included show that the bus fails to reach all but the first (Fort McHenry) and fifth stop
(Fulton & North Ave.) within the scheduled time constraints. The route class that is
calculated after the vehicle routing analysis is solved includes attributes of the route
generated. According to the route class layer, the entire route drive time was 69.12
minutes, at a distance of 55153.64 feet, or 10.45 miles. The route began at the first
stop at Fort McHenry at 4:08:00 PM, and reached the final stop at Sinai Hospital at
5:17:06 PM. However, the time window violations are small with a total violation
drive time of 23.11
minutes.
Since the time violations are small, they can be attributed to the drive time cost of the
distance driven and the cost of the turns. The route traveled to each stop is fairly
direct. This shows the route analysis routing the bus to each stop by using quickest
route. This intern represents the shortest distance since time is the impeding factor.
The violation times are fairly consistent along the route. The second, third, and
fourth stop violation times are under three minutes for each stop. While not ideal, in
Table 3 - Drive Time Violations without Traffic
29. 21
reality this is not extremely late. However,
the time violations propagate downstream as
discussed earlier. Even with the fifth stop
being reached on time, and the sixth stop
being less than a minute late, the route is not
able to make up the time difference on the
subsequent stops, and therefore reaches the
later stops farther behind schedule. The
differences between the violation times
show the amount of time lost or gained
while in route to the next stop. The results of the first route calculation show that the
bus route can stay on schedule fairly easily, and that there is potential for dynamic
bus routing to improve the results.
Routing Analysis Using K-Factor
The results of the first routing analysis with the K-Factor traffic parameter
included show a vast improvement over the control route. The route reaches all the
stops within the scheduled time constraints except for the fourth stop at Fayette &
Carey. The route class that is calculated after the vehicle route analysis is solved
includes attributes of the route generated. According to the route class layer, the
entire route drive time was 62 minutes, at a distance of 67838.53 feet, or 12.85 miles.
The route calculated using the K-Factor traffic attribute was about seven minutes
quicker than the route calculated without traffic. The route was also about two and a
half miles longer. This can be attributed to the route analysis avoiding higher traffic,
Figure 8 - Route without Traffic
30. 22
and thus higher cost road segments to minimize the time since it it’s the impeding
factor. The route began at the first stop at Fort McHenry at 4:08:00 PM, and reached
the final stop at Sinai Hospital at 5:10:00 PM. This falls within the time constraints
set forth by the schedule. All the stops are reached on time except the fourth stop.
However, the time violation for this stop is only 0.68 minutes. This violation is just
over 30 seconds, and is
almost negligible in a real
world scenario. This
reduction in total drive
time as well as the
reduction in the time
violations is due to the routing analysis avoiding road segments with high K-Factor
values. This intern represents the bus avoiding road segments with high traffic
densities. The dynamically routed bus route conserved time by avoiding traffic. The
dynamically routed bus conserved so much time that it arrived early at most of the
stops. If a bus arrives at a stop before the scheduled time, the bus must wait until the
scheduled time to depart and go to the next stop. This is represented by a weight
drive time. The bus waited at
all but two of the stops for a
total of 24.43 minutes. These
weight times represent the
amount of time saved by
dynamically changing the bus
Table 4 - Drive Time Violations with Traffic
Table 5 - Wait Drive Time
31. 23
routes to avoid traffic congestion. The wait times are largest at the stops that are
located near the downtown area. This is due to the re-routing of the bus early on in
the route. The route avoids a higher traffic area of southern Baltimore. This intern
saves time and causes the route to reach the second, third, and fifth stops early. The
results of the second route calculation show that the by dynamically changing the bus
route, the route can avoid traffic congestion and conserve the bus schedule times.
Figure 9 - Route with Traffic
32. 24
Chapter 5: Conclusion/Discussion
Conclusion
Based on the results of the second routing analysis, dynamic bus routing is
able to preserve the time window constraints of bus routes regarding arrival times at
individual bus stops. Based on the results of the analysis, the ability to change the
route to avoid road segments with higher traffic congestion allows the bus route to
make up time and reach the stops ahead of schedule. The first routing analysis shows
how bus scheduling delays propagate downstream. A bus that is delayed arriving at a
stop will be delayed arriving at a subsequent stop. This delay time will increase in
relation to stops further along the bus route. Through the research, this paper has met
the objectives that it set forth earlier. Based on the results of the routing analysis,
dynamic bus routing is a viable solution to mitigating traffic delays in public bus
transportation systems in urban areas.
Discussion
As with all research, there are areas for improvement. A main constraint of
dynamic bus routing is the ability to dynamically re-route the buses in real-time.
Getting traffic and re-routing information to drivers would be needed to make bus
routing truly dynamic. Real-time traffic and routing information would enable bus
drivers to change their routes regarding specific circumstances such as traffic, road
closures, traffic accidents, and/or construction. Change the route between two
individual stops may prove more useful than re-routing the entire route in certain
circumstances. The element of real-time dynamic bus routing needs to be explored
33. 25
further. A second constraint of dynamic bus routing involves the use of the K-Factor
traffic parameter. The K-Factor is a factor if increased volume at the 30th
peak hour
over a years-worth of Average Annual Daily Traffic (AADT). Even though the K-
Factor is effective and widely used in traffic engineering, for dynamic bus routing it
is far to general. The K-Factor represents a very long running average of traffic data,
and is more useful of a factor in determining street capacity and overall general traffic
flow and structure rather than a traffic indicator in real-time. On the fly traffic flow
data would be more useful in determining the viability and practicality of real-time
dynamic bus routing. However, this research does show that transportation planners
and operators must learn to think on their feet if they are to improve transportation
services reliability and efficiency in an ever growing urban environment.
34. 26
Appendix
Map 1 - Route without Traffic.................................................................................... 27
Map 2 - Route with Traffic......................................................................................... 28
Map 3 - K-Factor ........................................................................................................ 29
Map 4 - Study Area..................................................................................................... 30
Map 5 - Street Data Layer........................................................................................... 31
Map 6 - Bus Route...................................................................................................... 32
Map 7 - Bus Stops....................................................................................................... 33
42. 34
Bibliography
Larsen, A., O, Madsen, and M. Solomon. "Partially Dynamic Vehicle Routing-
Models and Algorithms." The Journal of the Operational Research Society (2002):
637-46. Print.
O'Toole, Randal. "Rapid Bus, A Low-Cost, High-Capacity Transit System for Major
Urban Areas." Policy Analysis 752 (2014): 1-16. Print.
Singpurwalla, Nozer. "Network Routing in a Dynamic Environment." The Annals of
Applied Statistics 5.2B (2011): 14007-424. Print.
Solomon, Marius, and Jacques Desrosiers. "Time Window Constrained Routing and
Scheduling Problems." Transportation Science 22.1 (1988): 1-13. Print.
White, Peter. "FACTORS BEHIND RECENT BUS PATRONAGE TRENDS IN
BRITAIN AND THEIR IMPLICATIONS FOR FUTURE POLICY."
International Journal of Transport Economics / Rivista Internazionale Di
Economia Dei Trasporti 36.1 (2009): 13-31. Print.
Yan, Shangyao, and Ching-Hui Tang. "An Integrated Framework for Intercity Bus
Scheduling Under Stochastic Bus Travel Times." Transportation Science 42.3
(2008): 318-35. Print.
Maryland Transportation Administration. Bus Stop Optimization Policy (Pilot):
Bus Network Improvement Project – Phase One Plan.
Maryland: 2014. Print.
