Professional Documents
Culture Documents
Wiley Encyclopedia of Operations Research and Management Science, edited by James J. Cochran
Copyright © 2010 John Wiley & Sons, Inc.
1
2 OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS
Strategic
characteristics Set of services
Network and route (routes), interchanges,
design and terminals
Fare structure and
budget constraints
Demand
Characteristics Fleet size requirement
Frequency setting
Service standards and Service frequency for
additional constraints Fleet sizing each time period
Route travel times for Timetable Departure and arrival
Tactical
each time period development times for buses
on each route
Operational
distributions
Vehicle schedules
Relief points
Control
Distribution recovery
Real-time control plans
blurred and the process is enriched by feed- modifying fares, and offering operational
back on previous decisions. Feedback is also variations on services such as deadheading,
employed in analyzing the impacts on subse- expressing, and short turning. In a two-way
quent stages of modifications to earlier ones. service, when demand for service is signif-
According to this scheme, the design and icantly imbalanced, a deadhead operation
analysis of a public transit system can be could prove beneficial. In such a case, transit
structured into four decision levels as follows. vehicles would operate normally, loading
Strategic decisions are those which are passengers, on the high-demand direction
difficult to modify later, at least for relatively and return empty (and as fast as possible,
lengthy periods. Examples of such decisions maybe using a different path) on the opposite
are the transit mode (e.g., BRT versus LRT direction. Such a scheme increases the supply
versus Metro); the type of vehicle to acquire where it is most needed while decreasing it in
(capacity); the structure of services to be the low-demand direction, allowing for a bal-
offered; and, at a lower level, the location anced occupation rate of the vehicles in both
of stations and stops along the routes. directions. An express service is a service that
Tactical decisions are those that hold over skips some predefined stops, offering a fast
the medium term and are revised anywhere trip to its passengers. A short-turning service
from monthly to half-yearly. They include associated with a certain route consists of
adjusting service frequencies at each period, providing a shorter service that operates only
OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS 3
in a section of the route with the purpose of addition, these decisions are influenced
increasing supply in its busiest section. by the institutional framework chosen
Operational decisions, such as drivers’ by the government, for example, dif-
shifts, and route and vehicle assignments, ferent agencies in charge of different
typically occur in the short term (daily basis). modes (e.g., buses and metro) or a sin-
Finally, control decisions define the sys- gle agency in charge of the complete
tem’s recovery response to the inevitable dis- system, such as transport for London.
ruptions that upset service schedules such as The answers to these questions prob-
absent drivers, vehicle breakdowns, services ably have significant impacts on the
running late, vehicle bunching, and so on. systems costs and the quality of service
In this article we discuss how operations faced by the users.
research provides answers to the problems 2. The transportation modes that will con-
confronting each type of decision. The text is figure the system (metro, light rail,
structured around the four decision-making BRT, and regular buses). For surface
levels: strategic or long term in the section modes, it is important to define the
titled ‘‘Strategic Decisions,’’ tactical or level of interaction that the system
medium term in the section titled ‘‘Tactical will have with other modes, partic-
Decisions,’’ operational or short term in ularly private cars. The more exclu-
the section titled ‘‘Operational Decisions,’’ sive is the transit right-of-way and the
and disruption response or real-time control smoother is the boarding and alighting
in the section titled ‘‘Real-Time Control.’’ process, the higher will be the vehicle
A deeper perspective of all these issues can speeds, generating significant benefits
be found in Vuchic [4]. in terms of level of service to users.
Also essential is the definition of the
STRATEGIC DECISIONS type of energy to be used by the vehi-
cles, since once decided it can be very
This decision level deals with questions relat- costly to change. Note that per passen-
ing to the formal design of a new and complete ger transported, the transit system is
transit system (a ‘‘greenfield’’ project) or the less polluting than cars; so it is intrin-
modification of an existing one. An example sically associated with a more sustain-
of a city designed and implemented from the able motorized urban mobility. Still, in
ground up is Brasilia, created in the mid- many cities the transit system as a
twentieth century, but it was initially mostly whole is a major actor in the generation
designed for cars. Other cities have faced the of pollution (atmospheric and acoustic).
challenge of drastically changing a transit 3. The type of route structure of the
system already in place [5]. In either sit- transit network. This decision is inti-
uation, strategic decision making needs to mately linked to the level of fare and
address the following questions: operational integration between the
routes. If there is no fare integration,
1. The market structure that will gov- many of the routes will tend to be
ern the system. How many operators designed to cross the entire area served
will participate in the system? Will so that most riders can reach their
it be based on private operators? Do destinations paying a single fare. Such
you expect operators to compete for the routes tend to be inefficient given that
passengers? What role will the govern- along most of their length, vehicle load
ment have? Different cities have tried factors are relatively low. This oversup-
different approaches, from full deregu- ply of underutilized vehicle kilometers
lation and fully privately operated to a can be reduced by adopting a sys-
centralized system run by the govern- tem design that ensures integration
ment. In between, there are centrally between routes. Typically, the design
operated systems with private opera- is a trunk-and-feeder system in which
tors such as Transmilenio in Bogota. In trunk services cover large distances
4 OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS
at high speeds and high vehicle load loaded vehicles), the disutility of trans-
factors while feeder services shuttle to fers, and in-vehicle travel time. Vehicle
and from the trunk lines, picking up load factors should also be considered
riders at their origin points and deliv- for their effects on passenger comfort.
ering them at their final destinations. In the case of high-frequency services
This setup significantly reduces fleet that are not operating to a set schedule,
and operating costs but forces users waiting time is assumed to be inversely
to transfer. It also requires a road proportional to frequency. The fare is
infrastructure and right-of-way that not included as a social cost since it
enables high operating speeds along is a mere transfer from one element of
transit corridors with stops or stations the system (the user) to another (the
that can efficiently handle transfers on vehicle operator) rather than the con-
a large scale. Varying levels of route sumption of a resource.
integration are observed in transit To attach cost figures to these impacts,
systems around the world, including the value of each of them—access time,
total competition among different waiting time, and penalties for trans-
vehicles on a single route, competition fers and overcrowding—must be deter-
among different routes, competition mined. These values greatly vary from
among different modes, and total fare user to user if they are estimated on the
and operational integration. basis of wage rates, but the use of a com-
mon value for all riders is recommended
4. The specific routes to be supplied. It is
so that the model is not biased toward
necessary to determine for each route
those with greater purchasing power
its actual path or course, the type of
(who highly value time). It is up to the
vehicle, and the service frequency. At
transit authority to define the value of
this decision level the analysis is typ-
time, but it should be noted that the
ically confined to the highest demand
higher is this value, the denser will be
periods of the day.
the network design and the more fre-
quent will be the services. It is therefore
In responding to these challenges, plan- an extremely important decision that is
ners attempt to identify a configuration that typically grounded on how significant a
maximizes social welfare, this being defined role the transit system is expected to
as the total benefits of trips undertaken less play in the city.
the social cost of the system in terms of 2. Infrastructure Costs. Different transit
resources allocated by society as a whole to technologies and services require
enable transit users to reach their chosen different levels of capital investment.
destinations. Of course, quite often this max- Putting a value on these costs necessar-
imization procedure is restricted to satisfy ily means defining the number of stops
some kind of budget constraint. Typically, or stations, the level of segregation of
this process involves building a model that the bus lanes, and the expenditures
represents overall generalized costs for a involved in building and equipping
fixed and known level of transit trip demand the corridor. The cost of installing any
specified by a trip matrix of travel between vehicle control equipment or traffic
different zones of the area served. The follow- signal priority systems must also
ing types of costs should be included: be included. Each of these decisions
directly impacts on the operating speed
1. Costs That Reflect Impacts on Users. of services.
Among the direct impacts of a trip are 3. Operation Costs. These typically
walking time (i.e., stop or station access include the cost of fleet acquisition and
time), waiting time at each stage of a unit operating cost per run for each
the trip (including added time when route (for a given type of vehicle). By a
waiting users are passed up by fully run, we understand the trip made by
OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS 5
a single vehicle from the beginning to Thus, if capacity constraints are included,
the end of a route. the design must analyze how users decide
4. Costs Imposed by Externalities. This their route when there is more than one
includes impacts on travel times of alternative, each with its own combination
other transportation modes (cars, of times, transfers, and fares. Models of
bicycles, and walking), impacts on user behavior need to take into account
accident rates, atmospheric and noise that riders normally place a higher value
pollution, and so on. on waiting than walking and more value on
walking than in-vehicle traveling [6]. They
The formulation should model demand as must also consider that these values greatly
a set of trips where each user chooses the best vary between users of different income
alternative available. In the absence of con- levels.
gestion or vehicle load (capacity) limits, such Incorporating all of these different ele-
a set can be identified simply by minimizing ments into a model is no easy task, and often
the social costs of the system. some are left out of preliminary design ver-
If congestion in the system is expected sions. Nevertheless, it is important that the
to be significant, however, with vehicle load model accurately represents the transit sys-
levels close to maximum capacity, the model tem so that it can generate detailed scenarios
will have to accommodate this added com- for comparing alternatives and supporting
plexity. A design that omits this factor and decision making.
merely minimizes total costs would wrongly Certain other indicators are useful in pro-
assume that riders spontaneously choose the viding general orientation for the various
alternative that is best for the system and decisions confronting transit planners. For
not just for them. Consider, for example, the example, it can be shown that the optimal ser-
simple case illustrated in Fig. 2 of a transit vice frequency (f ∗ ) of a route with no capacity
corridor with two lines in which their fre- or budget restraints is given by the following
quencies need to be determined. The corridor formula [7,8]:
has three nodes and two origin–destination
pairs (A–B, A–C).
Let us assume that line 1 is very fast along ∗ kcw q
f = , (1)
its entire length but has high operating costs co
while line 2 is slow but with low costs. Both
lines charge the same fare. Clearly, the fre- where cw is the value of wait time, q is the
quency on line 1 cannot be zero since it is the arrival rate of passengers to the system, co
only route supplying trips from A to C and, is the operating cost per run, and k is a
since it is offered, A–B passengers will also headway constant defined so that k = 1/2 if
want to take it. But if a capacity constraint headways are perfectly regular and grows
for line 1 were active, it could be cheaper as they become increasingly irregular. This
for the system to have A–B passengers to formula can be used to deduce the effect of
travel on the slower line than to increase fre- an abrupt decline in operating costs or a rise
quencies on the faster one, in which case the in demand on the optimal route frequency.
model would indicate that they wait for Line Note, however, that in developing countries
2, something they would not do of their own the service frequencies given by Equation (1)
accord. may be insufficient to meet demand. This is
because the value of operating costs tends to
be much higher than of wait time, implying
l2
co > > cw . In such cases, the frequencies will
l1 l1 have to be adjusted to ensure that demand
A B C on the most heavily traveled segment of each
route is satisfied.
Figure 2. Corridor with two O–D pairs served by The lower bound for the number of tran-
two lines. sit vehicles required during peak periods, n,
6 OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS
where Dv,k (t) is the cumulative number single driver) must be determined so
of departures of type v vehicles from that all tasks are assigned to a duty
terminal k at instant t, and Av,k (t) and each duty can be performed by a
is the cumulative number of arrivals driver. The possibility of drivers doing
of type v vehicles at terminal k at overtime at a higher salary is usually
instant t. The sum of these assign- considered in a CSP. Driver constraints
ments over all terminals and types of like maximum working time without a
vehicles gives the minimum required break, minimum break duration, and
fleet size for satisfying the entire maximum working time are typically
system schedule. This value can be added. Although CSP looks similar to
reduced through modifications to the the SDVSP in structure, it is much
route schedules such as marginally more complicated. The CSP has been
reducing frequencies, advancing or shown to be NP-hard. Wren and
delaying arrivals, changing assigned Rousseau [13] present an overview of
vehicle types for a given run, and different approaches for this problem.
repositioning vehicles from terminals Once the driver duties over the plan-
where more vehicles arrive than the ning horizon have been determined,
number departing for terminals with a duties are assigned to specific drivers.
shortage of arrivals. This repositioning At this stage, called the rostering
usually involves deadhead runs, but process, other constraints like rest
where demand exists it may also be periods, holidays, training, and so on,
traveled with passengers (creating new are considered. A survey on different
express services, for example). approaches for this problem can be
The vehicle scheduling process aims found in Odoni et al. [14].
at defining vehicle blocks consisting This sequential approach for transit
of feasible sequences of runs and systems is inherited from the airline
repositionings (deadhead vehicle trips) industry, where vehicles are optimized
for each vehicle between route ter- before drivers (crews in this case)
minals or terminals and the vehicle since vehicle-related costs have a
garage. The objective is to minimize a bigger impact on the total budget. In
combination of required fleet size and the case of transit systems based on
non-revenue time subject to satisfying buses, driver-related costs may be the
all scheduled runs. This problem of most significant. Therefore, transit
assigning vehicles to runs starting from systems may benefit from optimizing
a single depot is called the single depot drivers first and should do it taking
vehicle scheduling problem (SDVSP). full advantage of driver’s flexibility.
The SDVSP can be solved in polyno- This sequential approach to design a
mial time. For a detailed overview transportation system has other impor-
of the literature on the SDVSP, see tant limitations. Performance improve-
Desrosiers [12]. ments could be obtained if the problem
2. Driver Shift Scheduling. Once a block was solved simultaneously instead, but
has been assigned to each vehicle, they the complexity of the problem makes it
are initially divided into tasks that a challenge.
must be worked by a single driver. So Another drawback of the sequential
tasks start and end at relief points approach is that it assumes a determin-
of the network, where vehicles can istic system. However, uncertainties
exchange drivers. The problem of are present at all operation levels (e.g.,
assigning drivers to tasks is called the driver absenteeism, passenger demand,
crew scheduling problem (CSP). A CSP roundtrip durations, and vehicle fail-
is solved for each day of the planning ure) and should be considered in these
horizon. In the CSP, a minimum cost decisions (e.g., hiring extra drivers,
set of duties (sequence of tasks by a called extraboard, may allow operation
OPTIMIZATION OF PUBLIC TRANSPORTATION SYSTEMS 9