Sciencedirect

Available online at www.sciencedirect.
com
Available
Available online
online at www.sciencedirect.com
at www.sciencedirect.com
ScienceDirect
Available online at www.sciencedirect.com
Transportation Research
Transportation Procedia
Research 23 (2017)
Procedia 1025–1037
00 (2016) 000–000
Transportation Research Procedia 00 (2016) 000–000
www.elsevier.com/locate/procedia
Transportation Research Procedia 00 (2016) 000–000
OE*OUFSOBUJPOBM4ZNQPTJVNPO5SBOTQPSUBUJPOBOE5SBGGJD5IFPSZ
Tradingoffcosts,environmentalimpact,andlevelsofserviceinthe
optimaldesignoftransitbusfleets
a,∗ b
PabloL.Durango-Cohen
PabloL.Durango-Cohena,∗,ElaineC.McKenzieb
,ElaineC.McKenzie
aNorthwesternUniversity,2145SheridanRoad,Evanston60202,USA
aNorthwesternUniversity,2145SheridanRoad,Evanston60202,USA
bCambridgeSystematics,115S.LaSalle,Chicago60603,USA
PabloL.Durango-Cohen ,ElaineC.McKenzieb
a,∗
bCambridgeSystematics,115S.LaSalle,Chicago60603,USA
Abstract aNorthwesternUniversity,2145SheridanRoad,Evanston60202,USA
Abstract bCambridgeSystematics,115S.LaSalle,Chicago60603,USA
The development of a systematic framework to support the design of transit bus fleets is justified by the significant and long-lasting
The development of a systematic framework to support the design of transit bus fleets is justified by the significant and long-lasting
implications
Abstract associated with decisions to purchase transit vehicles, as well as by developments in fuel propulsion and battery
implications associated with decisions to purchase transit vehicles, as well as by developments in fuel propulsion and battery
technologies over the last 2 decades that have increased the options available to transit operators, and, in turn, the complexity of
technologies
The developmentover of
thea systematic
last 2 decades that have
framework to increased
support thethe options
design available tofleets
transit operators, and, in turn, the complexity of
assessing the corresponding tradeoffs. The need to evaluate theseof transit
tradeoffs bus
is, in part,is driven
justifiedbybythetheemergence
significant ofand long-lasting
environmental
assessing
implications the corresponding
associated tradeoffs.
with decisions The need
to purchase to evaluate
transit these
vehicles,tradeoffs
as well is, in part, driven
as by developments by the emergence
in fuel of
propulsion environmental
and battery
impact mitigation, i.e., emissions reductions, as a critical concern of transit operators and governments around the world.
impact mitigation, the
technologies i.e.,last
emissions reductions, asincreased
a critical theconcern of transit operators andoperators,
governments around the
theworld.
To addressoverthese concerns, 2 decades that have
we present an optimization options
model available
to support theto design
transit and,fleets
of transit bus in turn,
while complexityfor
accounting of
To address
assessing the these concerns,tradeoffs.
corresponding we present The anneed
optimization
to evaluate model
these to supportis,
tradeoffs theindesign
part, of transit
driven by busemergence
the fleets while of accounting
environmentalfor
costs, level-of-service requirements, and environmental impact. Methodologically, the work bridges applications of Economic
costs,
impact level-of-service
mitigation, i.e.,torequirements,
emissions and environmental
reductions, as alifecycle impact. of
critical concern Methodologically, the work bridges applications of Economic
Input-Output analysis conduct environmental assessment,transit operators
with seminal workand governments
in production around
economics.the world.
Input-Output
To address analysis
these to conduct
concerns, we environmental
present an lifecycle assessment,
optimization model to with seminal
support the workof
design in transit
production
bus economics.
fleets while accounting for
We apply the framework to support design of bus fleets consisting of 4 bus types differing in their fuel-propulsion technology:
We apply
costs, the framework
level-of-service to supportand
requirements, design of bus fleetsimpact.
environmental consisting of 4 bus types differing
Methodologically, the work inbridges
their fuel-propulsion
applications technology:
of Economic
ultra-low sulfur diesel, hybrid diesel-electric, compressed natural gas, and hydrogen fuel-cell. The 4 bus types were assessed in the
ultra-low
Input-Output sulfur diesel,tohybrid
analysis conduct diesel-electric,
environmental compressed
lifecycle natural gas, and
assessment, withhydrogen fuel-cell.
seminal studies
work The 4 buseconomics.
inconducted
production types were assessed in the
National Renewable Energy Laboratory transit bus evaluation and demonstration over the period 2003–2009.
National
We Renewable
apply the EnergytoLaboratory
framework support transitofbus
design bus evaluation
fleets and demonstration
consisting of 4 bus types studies
differingconducted
in their over the period 2003–2009.
fuel-propulsion technology:
The nominal problem herein is to minimize acquisition, operation and disposal costs. Constraints in the model are used to impose
The nominal
ultra-low problem
sulfur diesel, herein
hybrid isdiesel-electric,
to minimize acquisition,
compressed operation
natural andand
gas, disposal
hydrogen costs.fuel-cell.
Constraints
The in
4 the types
bus modelwereare used to impose
assessed in the
a minimum frequency of service, i.e., headway, and to ensure that route capacity satisfies passenger demand. Environmental
aNational
minimum frequency of service, i.e., headway, and to ensure that route capacity satisfies passenger demand. Environmental
impact is Renewable Energy
considered along theLaboratory transit bus evaluation
following dimensions: and demonstration
energy consumption, studiesofconducted
and emissions greenhouse over the period
gasses, 2003–2009.
particulate matter,
impact
The is considered
nominal problem along
hereintheisfollowing
to minimize dimensions:
acquisition, energy consumption,
operation and emissions of greenhouse gasses,areparticulate matter,
and nitrous oxides. Results show that fleet heterogeneity increasesand disposal
in scenarios costs.
whereConstraints in the model
demand fluctuates, i.e., peak used
vs. tooff-peak.
impose
and nitrous oxides.
aPerhaps
minimum frequency Results
of show that
service, i.e.,fleet heterogeneity
headway, and to increases
ensure thatinroute
scenarios
capacitywhere demand
satisfies fluctuates,
passenger i.e., peakEnvironmental
demand. vs. off-peak.
even more interesting, we show how the dual/shadow prices provide a (monetary) measure of the tradeoffs among level of
Perhaps
impact even more interesting, we show how the dual/shadow prices provide a (monetary) measure of thegasses,
tradeoffs among level of
service is
andconsidered along impact,
environmental the following dimensions:
and discuss how they energy consumption,
can be used to obtain and emissions
robust fleet of greenhouse
configurations. particulate matter,
service
and and
nitrous environmental impact, and discuss how they can be used to obtain robust fleet configurations.
�c 2016 Theoxides.
Authors. Results
Elseviershow B.V.that
Allfleet heterogeneity
rights reserved. increases in scenarios where demand fluctuates, i.e., peak vs. off-peak.
� c 2016 The
Perhaps even Authors. Elsevier B.V.
more responsibility
interesting, we All rights reserved.
© 2017
Peer The
review Authors.
under Elsevier B.V. ofshow
All rights how the
reserved.
the scientific dual/shadow
committee of prices provide
the 22nd a (monetary)
International measure ofonthe
Symposium tradeoffs among
Transportation andlevel
Trafficof
Peer review
service and under responsibility
environmental impact, ofandthediscuss
scientifichow committee
they can of used
be the 22ndto International
obtain robust fleetSymposium on Transportation and Traffic
configurations.
Peer-review
Theory. under responsibility of the scientific committee of the 22nd International Symposium on Transportation and Traffic Theory.
Theory.
�c 2016 The Authors. Elsevier B.V. All rights reserved.
Keywords:
Peer review Bus fleeting;
under Fuel-propulsion
responsibility of thetechnologies; Environmental
scientific committee design;
of the 22nd Input-Output
International models; Linear programming.
Symposium on Transportation and Traffic
Keywords: Bus fleeting; Fuel-propulsion technologies; Environmental design; Input-Output models; Linear programming.
Theory.
Keywords: Bus fleeting; Fuel-propulsion technologies; Environmental design; Input-Output models; Linear programming.
1.
1. Introduction
Introduction
The decision
decision to
to purchase
purchase transit
transit vehicles
vehicles has
has long-lasting
long-lasting implications
implications for
for the
the lifecycle
lifecycle costs,
costs, emissions,
emissions, and
and level
1. The
Introduction level
of
of service provided by a transit agency. Perhaps, because the variety of buses is relatively low, and because of
service provided by a transit agency. Perhaps, because the variety of buses is relatively low, and because of the
the
The decision to purchase transit vehicles has long-lasting implications for the lifecycle costs, emissions, and level
of∗ service provided
Corresponding author.by a transit
Tel.: agency. ; Perhaps,
+1-847-491-4008 because the variety of buses is relatively low, and because of the
fax: +1-847-491-4011.
∗ Corresponding author. Tel.: +1-847-491-4008 ; fax: +1-847-491-4011.
E-mail address: pdc@northwestern.edu
E-mail address: pdc@northwestern.edu
2214-241X c 2016author.
∗ Corresponding
� The Authors. Elsevier B.V. All; fax:
Tel.: +1-847-491-4008 rights+1-847-491-4011.
reserved.
2214-241X � c 2016 The Authors. Elsevier B.V. All rights reserved.
PeerE-mail
review
2352-1465 © under
address:
2017 responsibility
The of theB.V.
pdc@northwestern.edu
Authors. Elsevier scientific
All committee
rights reserved. of the 22nd International Symposium on Transportation and Traffic Theory.
Peer review under responsibility of the scientific committee of the 22nd International Symposium on Transportation and Traffic Theory.
Peer-review under responsibility of the scientific committee of the 22nd International Symposium on Transportation and Traffic Theory.
10.1016/j.trpro.2017.05.056
2214-241X � c 2016 The Authors. Elsevier B.V. All rights reserved.
Peer review under responsibility of the scientific committee of the 22nd International Symposium on Transportation and Traffic Theory.
1026 Pablo L. Durango-Cohen et al. / Transportation Research Procedia 23 (2017) 1025–1037
2 1Durango-Cohenand&McKenzie/TransportationResearchProcedia00(2016)000–000
increased complexity of managing heterogeneous fleets, the literature on bus fleeting is not extensive. However, as
technology has developed, buses have grown increasingly heterogeneous in a variety of important metrics, including
capacity, price, and operating characteristics. Recent and projected advances in alternative fuel and battery technolo-
gies have added an interesting dimension that will be relevant for the foreseeable future. The practical motivation for
our work, therefore, related to the complexity of decisions facing agencies in terms of assessing the relevant tradeoffs.
Methodologically, the work builds on Croft McKenzie and Durango-Cohen (2010), and bridges applications of
Input-Output (IO) analysis to conduct environmental lifecycle assessment (LCA), with seminal work in production
economics. In the latter, product design, production planning and scheduling problems are frequently formulated
as IO models with substitution, and subsequently analyzed and solved as linear programs (cf. Shephard (1953);
Hackman and Leachman (1989)). Indeed, and as described in Koopmans (1951), these types of problems are among
the first applications of linear programming. In addition to providing decision-support, the framework provides a
well-established approach to conduct sensitivity analysis, i.e., to evaluate the effect of perturbations in the inputs on
the results.
We apply the framework to support design of bus fleets consisting of 4 bus types differing in their fuel-propulsion
technology: ultra-low sulfur diesel, hybrid diesel-electric, compressed natural gas, and hydrogen fuel-cell. The 4 bus
types were assessed in the National Renewable Energy Laboratory transit bus evaluation and demonstration studies
conducted over the period 2003–2009. Data are also from the environmental LCA in Croft McKenzie and Durango-
Cohen (2012). The nominal problem herein is to minimize acquisition, operation and disposal costs. Constraints
in the model are used to impose a minimum frequency of service/maximum headway, and to ensure that capacity
satisfies passenger demand. Environmental impact is considered along multiple dimensions. Results show that fleet
heterogeneity increases in scenarios where demand fluctuates, i.e., peak vs. off-peak. Perhaps even more interesting,
we show how the dual/shadow prices provide a (monetary) measure of the tradeoffs among level of service and
environmental impact, and discuss how they can be used to obtain robust fleet configurations.
The remainder of the paper is organized as follows. In the next section, we provide a brief review of the relevant
literature. A formulation capturing the tradeoffs in designing bus fleets is presented in Section 3. Data used in the
present study are presented in Section 4. In Section 5, we consider a number of scenarios where the optimization
model is used to select bus fleets. Discussion of the results and conclusions of the study are presented in Section 6.
2. Background
Two streams of literature are related to the work presented herein. On the one hand, planning models aimed at
capturing the tradeoffs between costs and level-of-service, i.e., capacity and frequency/headway; and on the other
hand, environmental LCAs of transit vehicles.
In planning models, decisions to purchase a given type and number of buses are usually subordinated to routing
and scheduling problems. That is, the number of buses to satisfy demand is derived for a given level of service and
bus route/schedule, and under the assumption of homogenous bus type/technology. Hauer (1971); Navin (1979) are
pioneering studies. Vuchic (2005) is, perhaps, first to examine heterogeneity in bus capacity: either high or low
capacity, with corresponding price and operating cost differences. In addition to costs, tradeoffs between the buses
are explored along the following dimensions for peak and off-peak service: frequency/headway, passenger wait and
load factors. Hsu and Wu (2008) builds on the aforementioned models and proposes a fleet size model for number of
cars per train or BRT platoon. Other approaches to support bus fleeting include Khasnabis et al. (2003) and Peet et al.
(2009). The former studies optimal replacement schedules to meet long-term fleet needs. The latter proposes a tool to
allow transit operators to explore the tradeoffs between different bus technologies.
Environmental LCA of transportation and transit vehicles is an ongoing field of research. Chester and Horvath
(2009) has compiled what is, perhaps, the most comprehensive look at transit vehicles to date. Another substantial
evaluation of previous LCAs is presented in MacLean and Lave (2003). However, research that accounts for the
operating characteristics of alternative fuel vehicles is still in its early stages, in part, because deployment of such
vehicles is not widespread, and thus, (field) data are not widely available. Most studies on alternative fuel vehicles
have focused on automobiles and on tailpipe emissions. Findings on alternative fuels in transit vehicles are mixed.
In one of the first studies to relate environmental and economic costs, Johansson (1999) used lifecycle emissions
calculations to look at economic efficiency. He found that CNG buses can result in fuel savings of up to 35%. Hess
Pablo L. Durango-Cohen et al. / Transportation Research Procedia 23 (2017) 1025–1037 1027
1%VSBOHP$PIFOBOE&.D,FO[JF/TransportationResearchProcedia00(2016)000–000 3
(2007) reports that, although purchase price gaps were declining between CNG and diesel buses, higher labor and
maintenance costs are still significant. In a review of GHG-tailpipe emission and vehicle LCAs, Hesterberg et al.
(2009) concluded that these mixed results are, in part, attributable to the complexity and uncertainty involved in
measuring the lifecycle emissions associated with a vehicle.
Various approaches have been used to conduct the LCAs, with economic IO models, labeled EIO-LCAs, constitut-
ing an appealing one (Hendrickson et al., 2006). Rather than mapping processes in detail, e.g., chemical reactions, the
IO approach involves specifying the requirements or bill of materials of a product in terms of demand for economic
sectors such as transportation, construction, or financial services. The model, in turn, is used to compute the eco-
nomic activity and environmental repercussions associated with satisfying the given demand for the product. Because
all sectors represented in the economy are linked, there is no effective boundary on the scope of the analysis. The
number and diversity of EIO-LCAs has greatly increased since the late 1990s as a result of the methodology’s flexi-
bility, simplicity, and, importantly, the availability of tools and models to support the analysis. Examples can be found
in the fields of waste disposal (Kondo and Nakamura, 2004), transportation (Facanha and Horvath, 2007), and service
industries (Hendrickson et al., 2006). Although they have provided much insight, EIO-LCAs have been used almost
exclusively as descriptive tools. That is, EIO-LCA models have not been integrated into a prescriptive framework to
support decisions that arise during product/process design or (production) planning. To address this limitation, we
build on the model of Croft McKenzie and Durango-Cohen (2010), and use it to address the problem of designing bus
fleets.
3. Bus Fleeting Problem
We begin this section by describing the fundamental level-of-service tradeoffs inherent in the selection of transit
buses. We then formulate the fleeting problem as a linear program.
3.1. Tradeoffs between Frequency and Capacity
The fleeting problem consists of determining the number of buses needed to meet passenger demand at a prede-
termined level of service. The parameters of the problem fall under the categories of demand, level of service, bus
capacity, costs, and environmental impact. Demand corresponds to the number of passengers that needs to be served.
Capacity is the ability of each bus to serve demand. Level of service is decided by a policy to run buses at certain
intervals, which in turn determines the (average/maximum) time that passengers wait. Cost represents the cost to the
transit operator to own and operate buses. Economic tools, such as those described in Hauer (1971) can be used to
illustrate tradeoffs between these parameters. As an example, we consider the tradeoffs between service frequency
and capacity. Although others exist, we postpone a detailed discussion to Section 6.
Determining optimal frequency of service is a complex and ongoing issue for transit agencies. Although an infre-
quent service schedule may satisfy total passenger demand from a theoretical perspective, in practice it will also in-
crease passenger dissatisfaction and motivate the search for other transportation options. In fleeting models, frequency
is set by specifying a maximum headway. In low demand situations, this provides an incentive for the selection of
smaller buses, as they provide the same frequency of service at lower cost than larger buses of the same type. However,
smaller buses have less capacity per bus, and thus run at higher load factors. When load factor becomes critical (at
a certain number of passengers, or ostensibly when no one else is able to fit on the bus), either no more demand can
be served, or another bus needs to be added to the line. For the buses considered in the numerical example presented
herein, the tradeoffs between bus frequency and capacity are explored further in Figure 1.
The slope of each line indicates relationships between frequency (per hour) and the amount of passenger demand
that can be served. Horizontal and vertical lines represent example thresholds for minimum capacity necessary to
meet demand and maximum headway set by policy respectively. By observing the number of marks in each line to
cross each threshold, the number of buses required to meet a threshold is determined. For example, in Figure 1, “low”
demand can be served by all of the bus types, except HFC, with a frequency of 3 - 5 buses per hour (headway of 20 -
12 min), depending on type of bus. When a minimum frequency (maximum headway) constraint is added, the number
of buses required increases. Thus, these constraints work together to form the boundary on the minimum number of
buses. The frequency to capacity ratio is one aspect determining which constraints will be binding in an optimization
4 1%VSBOHP$PIFOBOE&.D,FO[JF/TransportationResearchProcedia00(2016)000–000
Frequency versus Line Capacity for Four Bus Types

10 6
8 7.5
Minimum headway
Frequency (buses/hr)
Headway (min/bus)
6 10
4 15
2 30
Off Peak Demand
Peak Demand
-
0 200 400 600 800
Line Capacity (spaces/hr)
Diesel Bus (C = 86) Hybrid Bus (C = 70) CNG Bus (C = 66) HFC (R) (C = 41)
Fig. 1. Line capacity and frequency analysis for the 4 bus types in this study. Adapted from Vuchic (2005). Additional information about the bus
types appears in Section 4.
problem. Costs of transit operations using smaller buses are more sensitive to capacity, while for an operator using
larger buses, costs are more sensitive to frequency/headway policy. Although this is a simplistic relationship, these
level-of-service tradeoffs illustrated in Figure 1 are the most fundamental part of the bus fleeting problem.
3.2. Bus Fleeting Model Formulation
A linear programming model to support the design of a bus fleet to serve a route with (round trip) service time T ,
demand per hour P.
�
Minimize: cr · n r (1)
r
Subject to:
� PT
sr · n r ≥ (2)
r
α
� T
nr ≥ (3)
r
h
⎡ ⎤
⎢⎢⎢ nr ⎥⎥⎥
⎢⎢⎢ − − − ⎥⎥⎥
� ⎢⎢⎢ ⎥⎥⎥
⎢ ⎥
Fk� · [I − Br ]−1 · ⎢⎢⎢⎢⎢ 0 ⎥⎥⎥⎥⎥ ≤ ek , ∀k (4)
r
⎢ .
⎢⎢⎢ . ⎥⎥⎥⎥
⎢⎢⎣ . ⎥⎥⎦
0
nr ≥ 0, ∀r (5)
where the decision variable, nr , represents the number of buses of type r to be purchased, cr is the equivalent
annual lifecycle cost, and (1) is the overall cost. sr is the capacity of bus type r. α and h are, respectively, the desirable
(maximum) load factor and headway. Thus, (2) and (3) ensure that service requirements are satisfied. Br is a bill of
materials matrix associated with bus type r (in terms of direct demand for activity across all economic sectors under
consideration), [I − Br ]−1 · [nr |0 · · · 0]� represents total direct and indirect economic activity associated with nr . Fk is a
vector with components representing environmental impact along dimension k, associated with economic activity in
each sector.1 Thus, each of the constraints in set (4) restricts total impact along dimension k to be below ek . Finally,
(5) impose logical restrictions on the decision variables.
To recap the contribution of the model herein, IO models for environmental LCA (e.g.: Croft McKenzie and
Durango-Cohen (2012)) are focused on term-by-term evaluation of the left-hand-side of Equation set (4). Due to
space limitations, we do not elaborate further, but do refer interested readers to Hendrickson et al. (2006) for a generic
discussion of the methodology, and to Croft McKenzie and Durango-Cohen (2010) for examples that include bills
of materials for substitutable alternatives. In contrast, the model in (1)–(5) integrates environmental impact with
economic and service consideration in the selection of nr . One of the reasons we choose not to pre-multiply the
parameters in the left-hand-side of (4) is to emphasize that one of the strengths of the modeling framework is that it
supports analysis of perturbations, due to technological developments, in the elements of B on the outputs. We also
note that the framework supports similar models such as minimizing environmental impact (across a given dimension,
e.g., greenhouse gas emissions) subject to budget constraints.
In the following sections, we consider numerical examples where the above model is used to design bus fleets for
a number of scenarios.
4. Data and Sources
The data used in this study were obtained from a series of recent demonstration studies on alternative fuel buses
conducted by the National Renewable Energy Laboratory (NREL), a subsidiary of the United States Department of
Energy (DOE). Data sources are summarized in Table 1. Using field data from bus operations poses challenges, but
has the potential to reveal insights. The use of a common and explicit set of assumptions to collect data allows for
reasonable comparisons across the studies. It also provides a benchmark for future data collection efforts.
Table 1. Data Sources: NREL Bus Demonstration Projects

Agency Location Year Bus Types
New York City Transit New York, NY 2004 Diesel, CNG, Hybrid
(Barnitt and Chandler, 2006)
Washington Metro Transit Washington, DC 2004 Diesel, CNG
(Chandler and Eberts, 2006)
Alameda-Contra Costa Transit Oakland, CA 2006 Diesel, HFC
(Chandler and Eudy, 2008)
SunLine Transit Agency Coachella Valley, CA 2008 HFC, CNG
(Chandler and Eudy, 2009b)
Connecticut Transit Hartford, CT 2008 Diesel, HFC
(Chandler and Eudy, 2009a)
In this study, we consider 4 types of transit buses: Diesel, Compressed Natural Gas, and Hydrogen Fuel Cell
(HFC). The parameters in the present study were generated by averaging over the data collected from each of the
demonstration studies. The scope of the life-cycle assessment and costs includes both bus manufacturing and oper-
ations. Operating characteristics such as capacity, fuel consumption, maintenance requirements; and manufacturing
characteristics such as size, technical components, batteries, power-train, and exhaust systems, are included. The
parameters corresponding to these characteristics were taken/adapted from the NREL studies, or are based on as-
sumptions inspired by the available data. They are summarized in Table 2.
1 A� represents the transpose of matrix/vector A. I is an identity matrix.

1030
1%VSBOHP$PIFOBOE&.D,FO[JF/TransportationResearchProcedia00(2016)000–000
Table 2. Parameters, Costs, and Environmental Impacts

Bus Types Diesel CNG HFC Hybrid
Parameters:
Bus Life (years) 15 15 15 15
Yearly Mileage 26,000 26,000 26,000 26,000
Capacity (pax.) 86 66 41 70
Costs:
Purchase Price ($1,000) 374 383 437 460
Fuel & Maintenance ($/mile) 1.71 1.62 1.58 1.57
LC Costs ($1,000) 777 792 837 857
Environmental Impact:
Manufacturing GHG (MT) 149 166 199 207
Operating GHG (g/mile) 3,287 2,540 2, 199∗ 2,453
LC Total GHG (MT) 979 807 754 826
Manufacturing NOx (MT) 2.7 3.1 3.8 4.0
Operating NOx (g/mile) 20.9 19.1 0.7∗ 15.6
LC Total NOx (MT) 8.0 7.9 3.9 7.9
Manufacturing Energy (TJ) 2.3 2.6 3.0 3.2
Operating Energy (TJ/mile) - - - -
LC Total Energy (TJ/mile) 2.3 2.6 3.0 3.2
Manufacturing PM (MT) - - - -
Operating PM (g/mile) 4.9 3.6 1.0 3.6
LC Total PM (MT) 1.2 0.9 0.2 0.9
∗
includes emissions from battery charging
5. Numerical Examples
We begin this section by outlining the scenarios considered for analysis. The following assumptions, some implicit
in the model, apply throughout (unless stated otherwise):
• Passenger demand is fixed and constant – it does not change based on day of the week or time of day.
• Buses are always able to travel at the same speed, as conditions of vehicle flow are not dependent on the number
of transit vehicles on the route or time of day.
• Boarding and alighting times are fixed and not dependent on how many passengers are boarding and alighting
at each stop.
• Maintenance, fueling, and crew issues do not affect the fleeting problem – that is, the daily bus service can be
completed without shift changes, fueling stops, or repairs.
• The length of the route is such that the yearly mileage of the bus is 26,000 miles and each round trip route takes
120 minutes.
• The infrastructure and personnel for each bus type is already in place.
• The life of the bus is 15 years.
• A discount rate of 6% is used.
• The maximum passenger load factor is the manufacturer’s specification of the number of seated and standing
passengers that can fit on the bus.
• Transit buses are purchased for a single line. Bus switching is not available.
• If passengers are not able to board the bus, they will wait for the next available bus.
5.1. Scenarios
The five scenarios below are intended to show the effects of a variety of parameter and threshold changes on the
optimal solution and to gain insight into the tradeoffs between buses, that can be applied to other fleeting problems.
In Scenario A, the fleet mix is optimized based solely on maximum headway and passenger demand. In scenarios
B-D, the four environmental impacts constraints are considered: greenhouse gasses (GHG), nitrous oxides (NOx),
particulate matter (PM), and total energy). In Scenario B, the tolerances are decreased proportionally for each impact,
resulting in an overall reduction of emissions. In Scenario D, NOx alone is restricted, while other environmental im-
pacts have modest tolerances. The remaining two scenarios, C and E, examine the effects varying passenger demands
and headway, respectively. Rather than representing a specific bus route, the objective is to highlight the framework’s
capabilities to capture the inherent tradeoffs.
Table 3. Description of Scenario Groups for the Bus Fleeting Problem

Scenario Description Environmental Passenger Demand LOS (Headway)
Constraints Constraints Constraints
A Baseline Scenario none yes yes, binding
B Low Impact GHG, NOx, PM yes yes

Total Energy, some binding
C High LOS GHG, NOx, PM yes max. headway,
Total Energy, some binding binding
D Low Emissions restricted NOx threshold yes yes, binding
E Increased Demand GHG, NOx, PM yes, binding yes

Total Energy, some binding
5.2. Numerical Results
The optimization model presented in Section 3.2 was used to select fleets for each of the scenarios in Table 3. The
results appear in Table 4. The first row shows lifecycle costs of each scenario, which include purchase and discounted
operating costs. The next box shows the fleet mix for each optimal solution. The third shows environmental impacts.
In some scenarios, levels of GHG and NOx reach the point where the respective constraints become binding. In these
cases, the shadow prices are reported. The final box summarizes other parameters. Again, when the headway and
demand levels reach the point where constraints become binding, the shadow prices are included.
5.2.1. Scenario A - Baseline

The first scenario focuses on servicing the passenger demand (4 pax per minute) with a maximum headway con-
straint (20 min). The optimal solution is to run 6 diesel buses, 20 minutes apart, at a lifecycle cost of $4.66m. Headway
(Equation 3), is the only binding constraint, and the load factor2 is 93%. The shadow price indicates that the decision
maker would be willing to pay (save) at most $38,869 to reduce (increase) the headway by one minute.
5.2.2. Scenario B - Fleeting with Environmental Constraints

In Scenario B, the environmental constraints are introduced by setting an allowable tolerance at the level in Scenario
A, and decreasing GHG and NOx till a minimum level is reached. Although the lifecycle cost does not change
significantly, the fleet mix changes to use CNG buses, as these buses have more favorable in terms of environmental
repercussions. The load factor increases slightly to 95%. The shadow price for the headway increases, indicating an
increased willingness to pay to relax this constraint. Furthermore, the shadow prices for GHG and NOx indicate that
the decision maker would be willing to pay to increase these tolerances as well.
2 The percentage of occupied sitting and standing places at the place where the demand is the highest.
Table 4. Summary of Results of the LCA-LP Bus Fleeting Model

Scenarios A B C D E
Lifecycle Costs $4.66m $4.67m $6.32m $4.71m $5.42m
Fleet Mix
Diesel 6.0 5.3 4.0 5.3 7.0
CNG 0.7 3.1 0.7
HFC 0.9
Hybrid
Environmental Impact
GHG (MT) 5,872 5,758 7,100 5,705 6,828
GHG Shadow Price ($/MT) $78 $78
NOx (MT) 48 47.9 60 45 56
NOx Shadow Price ($/MT) $10,301 $10,301 $14,678
PM (MT) 7 7 8 6 8
Total Energy (TJ) 14 14 20 14 16
Other Parameters
Headway (min) 20 20 15 20 17
Headway Shadow Price ($/min) $38,869 $43,718 $62,451 $44,740
Pax. Spaces/hour 258 251 292 241 300
Shadow Price ($/pax) $9,039
Load Factor 93% 95% 82% 100% 100%
5.2.3. Scenario C - Decreased Headway

In Scenario C, the maximum headway is decreased to 15 minutes. This initially makes the problem infeasible,
because the environmental tolerances are violated. Each constraint needs to be increased by 13 - 30%, with the
greatest increase in total energy in order to cover the amount of additional energy required to manufacture and use 8
buses instead of 6. The minimum cost solution is to use 4 diesel, 3.13 CNG, and .9 HFC buses, and the lifecycle costs
increase to $6.32m. The load factor actually decreases as more buses are used to carry the same demand, even though
some buses have less capacity. Again the shadow price for headway increases, and in this scenario reaches its highest
level at $62, 451.
5.2.4. Scenario D - Constrained NOx Emissions

Nitrous oxide pollutants a product of many transportation modes and are the building blocks of smog, acid rain,
and contribute to ozone depletion and global warming. Currently, National Ambient Air Quality Standards regulate
the amount of NOx that can be present in the outside air, and high traffic volumes contribute to non-attainment in
many areas. In Scenario D the primary motivation of the decision maker is to minimize the NOx pollutants from the
bus fleet.
When the NOx threshold is lowered, the optimal solution is to use more HFC buses, which has the lowest lifecycle
NOx emissions (see Table 2) However, due to the small passenger capacity of this bus, only 0.7 buses are replaced
with HFC buses before the load factor increase to 100%. Any further increase in demand or frequency will push the
solution over the NOx threshold. Shadow prices on multiple variables reflect the tight constraints in this scenario, and
can be used to value the tradeoffs that can potentially be made between different restrictions.
5.2.5. Scenario E - Increased Passenger Demand

How much can passenger demand be increased without increasing NOx emissions in the optimal solution? Scenario
E addresses this question. A strain is placed on capacity by increasing the passenger demand by 25% to 5 passengers
3 A “partial” or fraction of a bus can be obtained in the real world by running each bus less often than the others. For example, .5 of a bus means
that the bus in question would run half as often as the other buses. The remainder of the bus’s time could then be used to serve other needs, such as
running on alternate routes.
per minute. As demand is increased, more buses will be required. However, as the proportion of HFC buses is
increased, capacity will shrink. More HFC buses are needed to serve the same demand than any other type, increasing
overall emissions, thus HFC buses are not used in the optimal solution. Perhaps surprisingly, in this scenario the
optimal solution is to serve the route entirely with diesel buses. The minimum level of NOx is 56 MT. Even though
diesel buses have higher per-bus emissions, they are the only option with enough capacity to serve 5 passengers a
minute. Because no buses can serve the same amount of passengers as diesel buses without an increase in GHG or
NOx emissions, the minimization of these factors is also implied in this solution.
5.3. Sensitivity Analysis
As stated, one of the key advantages of the proposed framework is the capability to conduct sensitivity analysis as
described below.
5.3.1. Reduced costs

The reduced cost is a measure of the cost decrease necessary to bring a variable into the optimal solution. For
each variable not part of the optimal solution in Table 4, a reduced cost is given in Table 5. Reduced cost is one
method to determine the significance of the differences between different technologies, and the prices at which these
technologies will become part of the optimal solution, i.e. be cost effective. Depending on the scenario, the reduced
cost for each bus can range from < 5% to > 100% the cost of the bus. In the former case, this may be accomplished
by subsidizing the more environmentally friendly but less cost-effective technology for a small cost. In the latter case,
the cost serves as a barrier to entry, indicating that the technology is not a viable option for a particular situation.
Table 5. Reduction in Cost Necessary for Each Bus Type to Enter the Optimal Fleet Mix
Scenarios A B C D E
Diesel - - - - -
CNG $14,526 - - - -
HFC $59,298 $10,878 - $13,058 $466,072
Hybrid $49,360 $64,834 $66,046 $77,451 $223,991
5.4. Time of Day Tradeoffs
Although most research focuses on peak periods, off peak periods present an interesting challenge to modelers
and transit operators alike. Often, demand and acceptable level of service are quite different from the peak period.
Generally, the assumption is that buses should run less frequently in the off peak period. However, Jansson (1980)
found that social cost minimization requires more frequent bus service than is generally accepted, and makes the case
that in some scenarios the frequency of buses in the off peak period should match that of the peak period.
This section examines the problem of optimizing buses for the off peak period. Since demand is lower, it might be
feasible to run smaller, more efficient buses if they are available. Two scenarios are examined, one where the buses
are a subset of those available for the peak period, and one where the buses can be chosen for the off peak period
separately. Buses from Scenario C are chosen to illustrate the most possible tradeoffs. Secondly, the question of
minimum emissions in the off peak period is examined.
5.4.1. Description of Off-Peak Scenario Parameters and Objectives

Two cases are examined. In the first case, the fleet has already been optimized for the peak period with the solution
for Scenario C in Table 44 . The pre-optimized mix of buses includes 4 diesel buses, 3.1 CNG buses, and .9 HFC
buses. In the second case, the buses will be optimized for the off-peak period only; the solution is not constrained to
buses available in Scenario C.
4 Again we look at the single line problem only, so bus swapping is not allowed; the route must be run using selections from only this mix of
buses.
Unlike the peak period, both cases consider the optimal operational cost, that is, assuming that capital and fixed
costs are sunk; the buses are already purchased. Secondly, for each case, a model is run to minimize lifetime NOx
emissions by modifying the linear program so that the environmental constraint for NOx is moved to the objective
function. This formulation can be found in Croft McKenzie (2013). Although NOx emissions are the focus of this
paper, any environmental or level of service parameters can serve in the objective function, depending on the qualities
of the problem.
To model the off peak period, passenger demand is assumed to be 50% of during the peak period. Headways are
increased to 30 minutes (double that in the peak period), and then 40 minutes.
5.4.2. Time of Day Tradeoff Results

Figure 2 displays the optimal mix of buses, operational cost, lifecycle cost, and lifecycle NOx emissions for 4
off-peak models. The first set of columns shows results for a 30 minute headway, and the second set shows results
for a 40 minute headway. For each, minimized operational cost and minimized NOx emissions are modeled. The first
row shows the “Any bus” scenario, while the lower row shows the results if the bus types are constrained to those in
Scenario C.
Results show that even though allowing the choice from a wider set of buses can lower operational cost, it actually
causes an increase in lifecycle cost in all cases. The hybrid bus, which was not used in any of the peak scenarios,
is used due to its low operational cost when optimizing for the off-peak period alone. When minimizing NOx, the
lifecycle cost is greatly increased. However, when minimizing cost, even though the “Any bus” and constrained
scenarios have different mixes and operational costs, the lifecycle costs are similar (and lower in the constrained
scenario). Many costs of owning a mixed fleet, such as increased training, maintenance, and infrastructure costs, are
not included in this analysis, which would further increase the costs of running a mixed fleet.
30 Minute Headway 40 Minute Headway (+33%)
Minimize OP Minimize OP
Minimize NOx Minimize NOx
Cost Cost
1 .1
Choose Hy b
Any Bus 1 .9
Dsl
4 .0 5 .9
5.9 HFC
Hy b HFC
Operational Cost
($1000) $1588 $2336 $1253 $2336
Total Life Cycle
$3428 31.5 MT $4900 23.2 MT $2321 23.9 $4900 23.2 MT
Cost ($1000) / NOx
Choose From 0 .9 0 .9
Pre-Optimized HFC HFC 0 .9 0 .9

CNG
Mix of Buses
HFC
Only 2 .1
3 .1 3 .1
Dsl 2 .4
CNG CNG
Dsl
Operational Cost $1628 $1628 $1271 $1377

Total Life Cycle
$3208 28.0 MT $3208 28.0 MT $2345 23.9 $2593 22.5 MT
Cost ($1000) / NOx
Fig. 2. Optimal Off Peak Fleeting Strategies. Costs are in $1,000 and NOx emissions are in Metric Tons (MT).
6. Discussion, Conclusions, and Future Research
We present an optimization model to support the design of transit bus fleets while accounting for tradeoffs among
costs, level-of-service requirements, and restrictions on emissions and energy consumption. Specifically, the objec-
tive considered herein is to minimize acquisition, operation, and disposal costs. Constraints in the model are used
to impose a minimum frequency of service, i.e., headway, and to ensure that capacity satisfies passenger demand.
Environmental impact is considered along the following dimensions: energy consumption, and emissions of green-
house gasses, particulate matter, and nitrous oxides. To illustrate the framework, we consider scenarios of designing
bus fleets consisting of 4 bus types differing in their fuel-propulsion technology and ensuing design and operational
characteristics: ultra-low sulfur diesel, hybrid diesel-electric, compressed natural gas, and hydrogen fuel-cell. Data
for the study are from the National Renewable Energy Laboratory transit bus evaluation and demonstration studies
conducted over the period 2003–2009.
Methodologically, the work bridges applications of Economic Input-Output analysis to conduct environmental
lifecycle assessment with seminal work in production economics, where such analysis relies on linear programming
for the purpose of evaluation and selection among alternatives, as well as to support sensitivity analysis. From an
application perspective, the framework provides a structured approach to capture and assess potentially complex
tradeoffs in a variety of scenarios. Among the observations and insights that stem from our analysis, we note:
• First and foremost, the model highlights the sensitivity of efficient fleet mixes to problem specifications and to
the parameters considered in each of the scenarios.
For example, the mix of transit vehicles in the peak scenarios appear to be driven by the need to satisfy
passenger demand, i.e., diesel buses with the largest capacity appear in optimal solutions for all scenarios
considered. However, this is only part of the story, as the reduced costs suggest that small reductions/subsidies
of on the order of 5-10% of the purchase price of CNG and HFC buses would result in changes in the optimal
fleet mix, and consequently, in the optimal number of vehicles in 3 of 4 relevant scenarios.
In our analysis, modeling the off-peak scenario was a chance to examine different problem specifications –
alternative objective functions either seeking to minimize NOx emissions or operational costs –, or constraints
on the available alternatives – “Any Bus” vs. “Pre-Optimized” set. Generally, these scenarios favor the advanced
technology buses with lower operational costs, i.e., hybrid and HFC buses. However, when minimizing cost,
even though the “Any bus” and constrained scenarios have different mixes and operational costs, the lifecycle
costs are similar (and lower in the constrained scenario). Many costs of owning a mixed fleet, such as increased
training, maintenance, and infrastructure costs, are not included in this analysis, which would further increase
the costs of operating a mixed fleet. Therefore, even though the costs of running the buses optimized for the peak
period may look higher on the balance sheet, the overall costs of running this type of fleet will be minimized.
• Overall, HFC buses, which are generally known for their environmentally friendly operation, save emissions
only if the demand is low, because the emission per passenger are actually very high. Perhaps surprisingly,
diesel buses are some of the best performers in terms of emissions per passenger for all categories except PM,
and performed well in reducing overall lifecycle costs and emissions in a variety of scenarios.
• Jansson (1980) showed more frequent service, even in the off peak period, minimized social costs; however, a
corresponding increase in both capital and labor costs is incurred. Off peak analysis in this study shows that
smaller buses with more frequent service can allow a high level of service yet may not always cause large
increases in capital and operating cost, especially when demand is high and environmental thresholds are tight.
The limitations of the model, including restrictive assumptions that motivate additional research:
• While we use the model successfully to generate insights about tradeoffs in purchasing and operating transit
buses, directions in which the model can be updated/extended to reflect other (practical) considerations are
discussed below. A few of these directions are not unlike considerations that arise in any type modeling.
Formulating the decision variables as continuous rather than discrete simplifies the computation, provides
an established analytical framework to conduct sensitivity analysis, but leads to solutions that may not be
readily implementable. Similarly, there is an argument to be made that a multi-objective optimization model
would better capture the inherent tradeoffs in the problem.
Omissions in the objective function include the costs associated with deploying infrastructure to support
transit vehicles, increased costs associated with operating heterogeneous bus fleets, among others. Some of
these costs can be difficult to estimate. For example, they may include additional inventory carrying costs
for spare parts that cannot be pooled across an entire fleet. In any case, the optimal solution to the fleeting
problem provides a benchmark for the performance of alternative fleets that may be under consideration, i.e.,
the objective function and constraints of the model can be evaluated for arbitrary fleets and the results can be
compared to those obtained with the optimal fleet mix.
Relative to synergistic models in bus scheduling or in the design of transit networks, there is room to im-
prove the representation of demand and user costs in the model, as well as to extend the formulation to the case
of multiple bus routes. For example, rather than imposing maximum load factor and headways exogenously, it
may be possible to include comfort and waiting in a user cost function, and use it to optimize the aforementioned
parameters. Other possible improvements in the model might be to specify an endogenous demand function
that depends on the transit fare, or having travel times, i.e., dwell times, depend on the number of passengers.
Relative to models used to study the adoption of advanced technologies in other contexts, the model could
be improved by considering how demand or operating characteristics might evolve, and by formulating the
problem as a dynamic optimization model.
• The data used in the study are from field demonstration studies conducted 10-15 years ago. We chose these data
because they are publicly available, and because they were collected under explicit and consistent assumptions,
which allows for a more reasonable comparison among the buses. Even though, qualitatively, the tradeoffs
inherent in present-day buses are not radically different, it would be relevant to update the model parameters
to reflect recent technological developments in propulsion technologies, e.g., batteries and electric motors of
Hybrid buses, emissions control systems for diesel buses, hydrogen fuel cells, etc. We are not, however, aware
of efforts by the NREL or other organizations aimed at conducting such studies in a systematic fashion.
References
Barnitt, R., Chandler, K., 2006. New York City Transit (NYCT) Hybrid (125 Order) and CNG Transit Buses: Final Evaluation Results. Technical
Report NREL/TP-540-40125. National Renewable Energy Laboratory, U.S. Department of Energy.
Chandler, K., Eberts, E., 2006. Washington Metropolitan Area Transit Authority: Compressed Natural Gas Transit Bus Evaluation. Technical
Report NREL/TP-540-37626. National Renewable Energy Laboratory, U.S. Department of Energy.
Chandler, K., Eudy, L., 2008. Alameda-Contra Costa Transit District (AC Transit) Fuel Cell Transit Buses: Third Evaluation Report. Technical
Report NREL/TP-560-43545-1. National Renewable Energy Laboratory, U.S. Department of Energy.
Chandler, K., Eudy, L., 2009a. Connecticut Transit (CTTRANSIT) Fuel Cell Transit Bus: Second Evaluation Report. Technical Report NREL/TP-
560-45670-1. National Renewable Energy Laboratory, U.S. Department of Energy.
Chandler, K., Eudy, L., 2009b. SunLine Transit Agency Fuel Cell Transit Bus: Fourth Evaluation Report. Technical Report NREL/TP-560-44646-1.
National Renewable Energy Laboratory, U.S. Department of Energy.
Chester, M., Horvath, A., 2009. Environmental assessment of passenger transportation should include infrastructure and supply chains. Environ-
mental Research Letters 4, 1–8.
Croft McKenzie, E., 2013. An Input-Output Framework for the Inclusion of Environmental Impacts in the Design of Goods, Systems, and Services.
Ph.D. thesis. Northwestern University.
Croft McKenzie, E., Durango-Cohen, P., 2010. An input-output approach for the efficient design of sustainable goods and services. The Interna-
tional Journal of Life Cycle Assessment 15, 946–961.
Croft McKenzie, E., Durango-Cohen, P., 2012. Environmental lifecycle assessment of transit buses with alternative fuel technology. Transportation
Research Part D: Transport and Environment 17, 39–47.
Facanha, C., Horvath, A., 2007. Evaluation of life-cycle air emission factors of freight transportation. Environmental Science and Technology 41,
7138–7144.
Hackman, S., Leachman, R., 1989. A general framework for modeling production. Management Science 35, 478–495.
Hauer, E., 1971. Fleet selection for public transportation routes. Transportation Science 5, 1–21.
Hendrickson, C., Lave, L., Matthews, H., 2006. Environmental Life Cycle Assessment of Goods and Services: An Input-Output Approach. RFF
Press, Washington, DC.
Hess, D., 2007. What is a clean bus? Object conflicts in the greening of urban transit. Sustainability: Science, Practice and Policy 3, 45–58.
Hesterberg, T., Bunn, W., Lapin, C., 2009. An evaluation of criteria for selecting vehicles fueled with diesel or compressed natural gas. Sustain-
ability: Science, Practice, and Policy 5, 20–30.
Hsu, L., Wu, J., 2008. Fleet-size model for light rail and bus rapid transit systems. Transportation Research Record 2063, 34–42.
Jansson, J., 1980. A simple bus line model for optimisation of service frequency and bus size. Journal of Transport Economics and Policy 14,
53–80.
Johansson, B., 1999. The economy of alternative fuels when including the cost of air pollution. Transportation Research Part D 1999, 91–108.
Khasnabis, S., Bartus, J., Ellis, R., 2003. An asset management framework for state DOTs to meet transit fleet requirements. Transportation
Research Record 1835, 74–83.
Kondo, Y., Nakamura, S., 2004. Evaluating alternative life-cycle strategies for electrical appliances by the waste input-output model. International
Journal of Life Cycle Assessment 9, 236–246.
Koopmans, T., 1951. Activity Analysis of Production and Allocation: Proceedings of a Conference. John Wiley and Sons, New York, NY.
MacLean, H., Lave, L., 2003. Evaluating automobile fuel/propulsion system technologies. Progress in Energy and Combustion Science 29, 1–69.
Navin, F., 1979. Optimal urban bus size. Transportation Research Record 663, 74–76.
Peet, K., Partridge, E., Hale, M., Le, P., 2009. Future fleets: An applied model for bus fleet planning based on energy costs and impacts, in:
Proceedings of the 89th Annual Meeting of the Transportation Research Board.
Shephard, R., 1953. Cost and Production Functions. Princeton University Press, Princeton, NJ.
Vuchic, V., 2005. Urban Transit: Operations, planning, and economics.. John Wiley and Sons, Hoboken, New Jersey. chapter 3.

Sciencedirect

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Sciencedirect

Uploaded by

Copyright:

Available Formats

Available online at www.sciencedirect.

3. Bus Fleeting Problem

3.1. Tradeoﬀs between Frequency and Capacity

Frequency versus Line Capacity for Four Bus Types

3.2. Bus Fleeting Model Formulation

4. Data and Sources

Table 1. Data Sources: NREL Bus Demonstration Projects

1 A� represents the transpose of matrix/vector A. I is an identity matrix.

Table 2. Parameters, Costs, and Environmental Impacts

Table 3. Description of Scenario Groups for the Bus Fleeting Problem

B Low Impact GHG, NOx, PM yes yes

E Increased Demand GHG, NOx, PM yes, binding yes

5.2. Numerical Results

5.2.1. Scenario A - Baseline

5.2.2. Scenario B - Fleeting with Environmental Constraints

Table 4. Summary of Results of the LCA-LP Bus Fleeting Model

5.2.3. Scenario C - Decreased Headway

5.2.4. Scenario D - Constrained NOx Emissions

5.2.5. Scenario E - Increased Passenger Demand

5.3. Sensitivity Analysis

5.3.1. Reduced costs

5.4. Time of Day Tradeoﬀs

5.4.1. Description of Oﬀ-Peak Scenario Parameters and Objectives

5.4.2. Time of Day Tradeoﬀ Results

30 Minute Headway 40 Minute Headway (+33%)

Pre-Optimized HFC HFC 0 .9 0 .9

Operational Cost $1628 $1628 $1271 $1377

6. Discussion, Conclusions, and Future Research

You might also like