You are on page 1of 6

2015 IEEE International Conference on Systems, Man, and Cybernetics

Assessment of Bus Rapid Transit (BRT) Time Lags


Under Probabilistic Uncertainties

Renata Dantas, Jamilson Dantas, Renata Dantas Gabriel Alves


and Paulo Maciel Federal Institute of Education, Department of Statistics
Informatics Center Science, and Technology of and Informatics
Federal University of Pernambuco, UFPE Pernambuco (IFPE), Federal Rural
Recife, Brazil Campus Recife, Brazil University of Pernambuco,
{rcspd, jrd, prmm}@cin.ufpe.br renatadantas@recife.ifpe.edu.br UFRPE Recife, Brazil
gabriel.alves@deinfo.ufrpe.br

Abstract—Large cities face growing mobility problems, due to This paper submits an evaluation method for quantifying
the major traffic jams that result from high numbers of vehicles the probability metrics of a BRT system. This method consid-
on the roads. In response, city and national governments have ers the system from the vehicle perspective, and the results pro-
invested in alternative means of urban passenger transit, such as vide important indicators for improving BRT system planning.
subways, trains, as well as Bus Rapid Transit (BRT). This article A CTMC model is created to describe the system and calculate
aims to analyze the BRT system, by attempting to calculate the the probability of reaching a particular destination at a given
probability of reaching a destination at a specific time, thereby
providing a tool that can be employed to improve the system and
time. From this model, two case studies are developed: the first
increase passenger confidence in it. To this end, a Continuous employs existing data from the literature to serve as a data
Time Markov Chain (CTMC) model is proposed to represent source for model verification; after achieving verification of
the bus stations and compute the probability metric for arrival the model, the second case study employs random information
at the destination within the specified time frame. The model from an actual BRT system in order to provide information for
allows a mathematical function to calculate the probabilities for improving the planning of that system.
the corresponding architecture. Two case studies were conducted
in order to verify the model and illustrate its potential value in
the planning of BRT systems. There are works in the literature which propose evaluation
techniques for BRT systems. Lopez et. al employ Stochastic
Keywords—Bus Rapid Transit (BRT), CTMC, Probability.
Petri Nets (SPN) to rearrange the stations of a BRT system
[4]. This rearrangement is undertaken in order to improve the
I. I NTRODUCTION system; and is based on passenger displacement time in the
Transmilenio BRT of Bogota, Colombia. The Petri net was
Rapid urbanization and growing traffic volumes create built according to stations on a specific route, and changes
problems such as congested roads and difficult mobility. It were made in the model to propose improvements. However,
is a recurrent theme in large cities that much time is wasted the work does not develop specific metrics that could be
by citizens moving between locations, whilst taking children employed in system planning. In [5] the authors address urban
to school, or going to and returning from work, as well as transport modeling with the Geographic Information System
performing all the other necessary daily activities that eventu- (GIS) method. This methodology considered the topographic
ally become troublesome for those living within a metropolitan and geographic aspects in order to analyze the aspects of time
area [1]. and space.
Beside the impact in terms of time wasted, heavy traffic
also creates environmental problems, such as noise pollution The current work offers a different approach to those
and excessive greenhouse gas emissions. There is also a described above. Here the modeling of the system is proposed
negative economic repercussion; the cost of congestion in the through the employment of CTMC, and a probability metric
U.S. in 2012, measured in terms of lost working hours and is developed for destination arrival. Additionally, study results
extra fuel consumption, was estimated at $120 billion [2]. are verified through a mathematical model, thereby creating a
tool which is directly relevant to arrival time planning in BRT
As an alternative to heavy traffic scenarios, governments
systems.
have invested in improvements to public transport systems,
in an attempt to create quality systems that enable citizens
to migrate away from private vehicle use, thereby reducing The remainder of the paper is organized as follows: section
the number of vehicles on the roads. A solution that has II introduces the fundamental concepts of BRT systems and
shown satisfactory results is Bus Rapid Transit (BRT) [3]. CTMC modeling; section III presents the proposed architec-
BRT operates on exclusive routes, with an off-board method ture; section IV discusses the CTMC model; section V presents
of charging customers, so the promise of greater reliability and the case study and highlights the results; and section VI draws
speedier transit make it an attractive proposition. some conclusions and indicates the direction of possible future
work.

978-1-4799-8697-2/15 $31.00 © 2015 IEEE 1273


DOI 10.1109/SMC.2015.227
II. F UNDAMENTAL CONCEPTS distribution for the initial state, it is possible to determine the
probability of being at any state at any time instant [12].
This section presents a summary of the concepts necessary
for understanding this work, and also provides an overview of This most general definition of a Markov process can
the performance model in the BRT system. be adopted for special cases. In the current paper the focus
is specifically on discrete state spaces and on continuous-
parameter Markov processes, and therefore this work deals
A. Methods and models for dependability evaluation
mainly with Continuous Time Markov Chains (CTMC), as
System dependability can be understood as the ability to defined below:
deliver a specified functionality that can be justifiably trusted A CTMC is a state-based model type, that represents the
[6].An alternate definition is “the ability of a system to avoid system behavior through states, and the occurrence of the event
failures that are more frequent or more severe, and outage is denoted by the state transition label. Labels can be proba-
durations that are longer than is acceptable to the user” bilities, rates or distribution functions. These models allow the
[6]. Dependability encompasses measures such as reliability, representation of more complex relationships between system
availability and safety [6]. components, such as dependencies involving subsystems and
Dependability is an important property for urban traffic resource constraints [9].
systems, since to be attractive to users they must provide In a CTMC model, the probability of being in one place
highly available and reliable services. BRT is an example of at any given time is calculated by Equation 1[11]:
a customer-orientated transit system designed to deliver fast,
comfortable and cost-effective urban mobility. However, as
traffic congestion increases, the dependability of urban traffic dP (t)
systems is becoming more important and at the same time P (t) = = QP(t) (1)
dt
more difficult to achieve [7].
Various model types may be applied to the analytical Where, P (t) is a matrix of the transition probability, t is
evaluation of dependability. Reliability block diagrams, fault the time and Q is a matrix infinitesimal generator.
trees, stochastic Petri nets and Markov chains have been This study is particularly concerned with a special type of
employed to model fault-tolerant systems and to evaluate Markov chain, called an absorbing Markov Chain. A state Si
various dependability measures. These model types differ from of a Markov chain is called absorbing if it cannot be exited
one to another in terms of modeling power [8]. They may be from (i.e., pii ), and a Markov chain is considered absorbing if
broadly classified into combinatorial and state-based models it has at least one absorbing state, and if from every state it is
[9]. State based models may also be referred to as non- possible to reach an absorbing state. In an absorbing Markov
combinatorial, and combinatorial models may be referred to chain, a state which is not absorbing is called transient [13].
as non-state based models. In some special cases state-based
analytic models can be solved to derive closed-form answers, C. BRT system
but generally a numerical solution of the underlying system of
equations is necessary [9]. As depicted in Figure 1, a BRT system can be viewed from
three different perspectives. Refer to performance character-
Combinatorial and state-space models may be hierarchi- istics together with the individual elements which determine
cally combined to achieve the best of both worlds. For how a useful system is generated. The Major Elements of
example, a reliability block diagram may be employed to a BRT are represented here, together with their respective
represent the dependability relationship between independent features and attributes: dedicated busways; stations and ve-
subsystems, while detailed or more complex fail and repair hicles; fare collection system; Intelligent Transport System
mechanisms are modeled with Stochastic Petri Nets (SPN) (ITS: fundamentally the creation of intelligent traffic lights);
[10]. This approach enables the representation of many kinds as well as the operation planning service and operation plan.
of dependency between components, and avoids the established System Performance refers to the following attributes: travel
issue of state-space explosion when dealing with large systems time savings, identity and image, safety and security, and
[10]. capacity.System Benefits refer to the various benefits a BRT
affords, including: high passenger levels; transit-supportive
B. Markov chain land use development; environmental quality; capital cost ef-
fectiveness; operating efficiency; and many other benefits. This
Markov processes constitute a special sub-class of stochas- three perspective structure suggests the relationships between
tic processes, while the latter can be considered as a gener- the major elements, system performance, and system benefits.
alization of the concept of random variables. In particular, a
stochastic process relates the elements of a possibly infinite Wright (2002) describes the main features of BRT systems
family of random variables. A series of random experiments as: segregated busways; rapid boarding and alighting; clean,
can thus be taken into consideration and analyzed as a whole secure, and comfortable stations and terminals; efficient pre-
[11]. boarding fare collection; effective licensing and regulatory
regimes for service operators; clear and prominent signage and
The main characteristic of Markov chains is that their real-time information displays; transit prioritization at intersec-
stochastic behavior is described by transition probabilities of tions; modal integration at stations and terminals; clean bus
the form P [X(tk+1 ) = x | X(tk = x)] for all state values x, technologies; sophisticated marketing identity; and excellence

x and tk ≤ tk+1 . Given these transition probabilities and a in customer service [7].

1274
Fig. 2: Architecture proposal.

which details travel times and from these calculates system


reliability, was employed to verify bus arrival probability.
The architecture proposed here considers a compound BRT
system of two central stations and six intermediate stations
(see Figure 2). This architecture was designed according to
real world BRT systems, whose primary objective is to link
suburbs to center, including within this parameter numerous
stops which attempt to meet passenger demand.

IV. P ROPOSED M ODEL


Fig. 1: Presentation of BRT system (Adapted from [14]). This section introduces the proposed CTMC model and
also discusses the validation of results by comparison with
real data extracted from [18].
Transport systems consider four specific performance mea-
sures: Mobility refers to the ability of the system to facilitate A. CTMC Model for bus arrival probability
efficient movement of people and goods; Reliability refers to
the ability of system users to predict the time it takes to make To create the CTMC model (3) that represents the BRT
trips on the system; Accessibility refers to the ability of the system, and perform the probability analysis, the following
system to connect people to desired destinations according to crucial factors were considered: station, time, and number of
the geographical distribution of the residential population, em- vehicles in the system. The model can be described with
ployment centers, and other service or recreation opportunities; a Hypoexponential distribution, which is the generalization
and Safety refers to the ability of system users to reach their of Erlang distribution, but having different rates for each
destination safely on any given trip [15]. transition [19].This can also be described by Equation 2 [11].

This paper considers the factors described above for model


building and analysis, highlighting the System Performance
criteria as the means to achieve the goals that provide the
essential differential for a BRT system, and therefore the
primary focus of the work. Although the Major Elements are
not directly considered, they are an important factor underlying Fig. 3: CTMC model for bus arrival probability.
the waiting time data and system displacement. System Benefits
are represented as the positive achievements of the work, in
terms of the possible consequences for the principal objective,
which is development planning. k

P DF = ai λi e−λi t , t > 0,
i=1
III. A RCHITECTURE P ROPOSAL k
(2)
 λj
The main objective of the work is to analyze the probability whith, ai = ,1  i  k
λj − λi
of a bus arriving at a final destination along a specified pathway j=1,j=i
over time. To solve the search problem, a baseline was con-
structed which united the description of the BRT system with Where for the Probability n Density Function (PDF),
the the metric definitions for analysis. This baseline was taken α=(α1 ,α2 ,α3 ,α4 ,...,αn ) with i=1 αi = 1, λi is the rate and
from the BRT descriptions found in [16], [17], which consider t is the time.
all the fundamental characteristics, components, structures,
performance evaluation criteria, and system benefits of a BRT The model of Figure 3 enables the obtainment of a closed-
network. form equation for calculating the Mean Time To Absorption
(MTTA), given by the average calculus in hypoexponential
From the baseline the high-level CTMC models were distribution (see Equation 3),employing the same parameters
built, and from these were derived the analysis and evaluation as Equation 2. The data concerning probability of bus arrival
models. The scenario described in the Changzhou BRT [18], at the destination and travel times at different times of the day

1275
on a daily journey were extracted from [18] and adapted to TABLE II: Input parameters for each peak time.
the proposal.
Average Waiting Average time on Value
Period of the day Parameters
Time (min) the vehicle (min) (min)
k λtt
 1 Early Time
λttw 1 to 6

1.19
3.26
3.26
3.26
4.45
MTTA = (3) λtt – 3.07 3.07
λ
i=1 i
AM Peak Time
λttw 1 to 6 0.31 3.07 3.38
λtt – 2.70 2.70
Inter-Peank Time
λttw 1 to 6 0.68 2.70 3.38
Table I describes the travel times for five different intervals λtt – 2.48 2.48
PM Peak Time
λttw 1 to 6 0.94 2.48 3.42
in particular route [18]. The travel ranges are: Early, AM Peak, λtt – 3.10 3.10
Evening Time
Inter-Peak, PM Peak and Evening. The maximum journey time λttw 1 to 6 0.96 3.10 4.06
for each travel time interval are considered as the sum of wait
time and travel time, where, the origin stop headway (Hi) is
the cumulative wait time for all stations and, In-vehicle travel
time (Ti) is the cumulative travel time in a pathway.

TABLE I: BRT system information from [18].


Origin Stop In-Vehicle Max. Journey Time
Trip
Headway (Hi) min Travel Time (Ti) min (Hi + Ti) min
Early 7.15 22.80 29.95
AM Peak 1.87 21.52 23.39
Inter-Peak 4.07 18.88 22.95
PM Peak 5.65 17.35 23.00
Evening 5.75 21.70 27.45

Fig. 4: Destination arrival probability.


The current study has seven stations along a pathway, and
so the dataset in Table I can be subdivided to represent each
station, at their respective peak time. Thus, Table II illustrates
the dataset for each peak time. The λttw can be subdivided
Another factor that can be considered is the probability of
from 1 to 6 stations. The transition λtt has no wait time, since
bus arrival within a specific time frame. For example, during
this is the starting lane, and it is assumed that the bus leaves
the Inter-Peak period the mean maximum journey time is 22.95
on time.
minutes (Table I). Therefore for a time frame of 25 to 35
minutes, as illustrated in Figure 5, there is an 80% probability
V. C ASE S TUDIES of the BRT vehicle arriving at its destination.
Two case studies were performed to demonstrate the suit-
The behavior during AM Peak and PM Peak is similar
ability of the model. The first study makes a comparison
to Inter-Peak,where probability of arrival within the 25-35
between the results obtained from the model with a simple
minutes interval is approximately 80%. The Early and Evening
study of the system given in [18]. In the second study a
intervals have probability of arrival rates of around 60%.
scenario is illustrated based on information obtained from a
BRT system currently being implemented in the city of Recife,
North-East Brazil.

A. Case Study I
Based on the dataset taken from [18] as given in Table
II, the probability of vehicle arrival time in the final lane in
a particular time period was calculated. Figure 4 illustrates
these probabilities. It can be seen that Early and Evening
demonstrate the worst behavior. This probably occurs as a
result of the fact that fewer buses are released at these times,
and the bus takes a longer route, thereby taking longer to
reach its destination. The AM peak, Inter-peak, and PM peak,
demonstrate practically identical behavior, reflecting the fact
that at peak times the number of buses on the route is greater,
given the increased demand, which results in a shorter arrival Fig. 5: Probability of arrival in a range.
time.
This relationship indicates that a transit planning process
is a fundamental requirement to assist in reducing the journey Table III compares the Mean Time To Absorption as
time by improving the probability of arrival at the destination, calculated by equation 3 with the actual data extracted from
in other words ensuring that the passenger arrives at his [18]. The very low figures for mean error suggest that the
destination on time. model is accurate and cannot be refuted.

1276
TABLE III: Average Error for Mean Time To Absorption. words, increasing the bus speed by a factor of 2 does not mean
that the bus will arrive in half the time.
Max. Journey Time Mean Time To Mean
Trip
(Hi + Ti) Absorption (by model) Error
Early 29.95 29.96 0.00033
AM Peak 23.39 23.39 0.00005
TABLE V: Relationship Between Mean Time To Absorption.
Inter-Peak 22.95 22.95 0.00011
Relationship Mean Time to Relationship
PM Peak 23.00 23.05 0.00200 TRIP
Between Trip Absorption (MTTA) Between MTTA
Evening 27.45 27.45 0.00080 23 Km/h 1 31.78 1.00
31 Km/h 1.35 24.72 0.78
39 Km/h 1.70 20.55 0.64
B. Case Study II
In Figure 6, which illustrates the probability of arrival at
Since there were no grounds for refuting the model, a a given time (t), it can be seen that for a time interval of 20
second case study was implemented based on a system cur- minutes the probability rises from less than 0.03 at 23 km/h to
rently being implemented in the city of Recife, North-East almost 0.5 at 39 km/h. This categorically confirms the impact
Brazil. Data was sourced from [20] regarding the number of of bus speed on arrival probabilities. Another important point
stations planned for the operation (22) and the average distance to note is that at a time interval of 40 min, the probability of
between them (500 meters). arrival at a vehicle speed of 23 km/h is 0.87, whilst at speeds
The scenario also took into account the average speed of 31 and 39 km/h, the probability is already at 0.99.
of the vehicle according to [21], which states that efficient
BRT systems have approximate average speeds between 23
and 39 km/h. Another important factor considered in the
scenario was the average stop time for passenger boarding and
disembarking. In this case the figure adopted was 22 seconds,
which is the given stop period for an articulated four-door
vehicle [21].
With this data three scenarios were examined, relating
to the minimum speed (23 km/h), the median average (31
km/h) and the maximum (39 km/h), hereafter referred to as,
respectively, scenario 1, scenario 2, and scenario 3. Table
IV,gives transition times for each scenario. Note that, as in the
previous study, there is no wait time (λtt ) at the first station
since it was assumed that the bus leaves on time.

Fig. 6: Destination arrival probability.


TABLE IV: Input parameters for each scenario.
Average Waiting Average time on Input Value
Scenarios Parameters
Time (min) the vehicle (min) (min) If the natural assumption is taken that a high probability
λtt – 1.304 1.304 of arrival within a given time period is perceived as having a
23 Km/h
λttw 1to21 0.220 1.304 1.524
λtt – 0.968 0.968 major impact on passenger confidence in the system, then it
31 Km/h
λttw 1to21 0.220 0.968 1.188 is important to establish an acceptable range. In this case, a
λtt – 0.769 0.769
39 Km/h
λttw 1to21 0.220 0.769 0.989 range of between 21 and 32 minutes might be considered as
an acceptable journey time to ensure customer satisfaction.
With this model the mean time to absorption can be
measured, representing the time that the BRT takes to get to
the last station. This is a prime item of information for BRT
system planning. In this case, as can be seen in Table V, the
absorption time, which has a direct influence on the probability
of arrival at a given time, is inversely proportional to the speed.
There is another point to make regarding the correlation
between absorption time and the increase in average speed:
the proportional increase in speed is greater than the decrease
in arrival times. In Table V it can be noted that scenario 1 at
23 km/h has an MTTA of 31.78. In scenario 2, the speed has
increased by approximately 35% (from 1.00 to 1.35), whereas
the decrease in MTTA is 22% (0.78 of the MTTA in scenario
1). In the third scenario the speed increases by 70% (from 1.00
to 1.70) in relation to scenario 1, while the decrease in MTTA Fig. 7: Probability of arrival within the required range.
is 36% (0.64 of the MTTA in scenario 1). Thus, it can be noted
that although the increase in speed is inversely proportional to According to these parameters, as illustrated in Figure 7,
the decrease in MTTA, there is no direct correlation. In other the strong impact of speed on arrival probability is clear; at

1277
23 km/h the probability of arrival within this frame is just 0.2, [3] J. C. Muñoz and D. Hidalgo, “Workshop 2: Bus rapid transit as part
whereas at the other end of the speed scale, at 39 km/h, the of enhanced service provision,” Research in Transportation Economics,
vol. 39, no. 1, pp. 104–107, 2013.
probability of arrival is 0.9.
[4] D. Lopez, A. Triana, and H. Chamorro, “Simulation model of public
This case study emphasized speed as the relevant factor transportation system using multiagent approach by means of petri
for guaranteeing destination arrival within the prescribed time nets: Bogotá study case,” in Robotics symposium, 2011 IEEE IX Latin
American and IEEE Colombian conference on automatic control and
frame. A very significant advantage of a BRT system is industry applications (LARC). IEEE, 2011, pp. 1–6.
that since it operates on an exclusive route and is therefore [5] P. Thorlacius, H. Lahrmann, and A. Pittelkow, “Time-and-space mod-
not influenced by traffic levels, reliable information regarding elling of public transport systems using gis,” Trafikdage på AUC, 1998.
travel times can be passed on to the passenger, thereby creating [6] A. Avizienis, J.-C. Laprie, B. Randell, and C. E. Landwehr, “Basic
a feeling of trust in the system which attracts more customers concepts and taxonomy of dependable and secure computing,” IEEE
to the service. Thus, through a combination of factors such as Trans. Dependable Sec. Comput., vol. 1, no. 1, pp. 11–33, 2004.
average speed and number of vehicles on the route, the system [7] L. Wright, “Bus rapid transit,” 2002.
can be productively employed to provide a mass means of [8] M. Malhotra, “Power-hierarchy of dependability model types,” IEEE
transport, creating better mobility conditions in the cities by Trans. on Reliability, vol. 43, no. 2, pp. 493–502, Sept. 1994.
decreasing the number of vehicles on the road and ensuring [9] P. Maciel, K. S. Trivedi, R. Matias, and D. S. Kim, “Dependability
modeling,” in Performance and Dependability in Service Computing:
that passengers can confidently predict their travel times. Concepts, Techniques and Research Directions. Hershey: IGI Global,
2011.
VI. C ONCLUSIONS [10] J. Dantas, R. Matos, J. Araujo, and P. Maciel, “An availability
model for eucalyptus platform: An analysis of warm-standy replication
This paper addressed the fundamental factors involved in mechanism,” in Systems, Man, and Cybernetics (SMC), 2012 IEEE
BRT systems, employing CTMC models as a means to foster International Conference on. IEEE, 2012, pp. 1664–1669.
system analysis and improve the planning process, particularly [11] G. Bolch, S. Greiner, H. de Meer, and K. S. Trivedi, Queueing
in regards to arrival probability and reliability. The models networks and Markov chains: modeling and performance evaluation
allow a mathematical function to calculate the probabilities of with computer science applications. John Wiley & Sons, 2006.
a corresponding architecture. [12] C. G. Cassandras and S. Lafortune, Introduction to discrete event
systems. Springer Science & Business Media, 2008.
The employment of case studies endorse the importance of [13] C. M. Grinstead and J. L. Snell, Introduction to probability. American
models for planning the system: they can capture information Mathematical Soc., 1997.
concerning destination arrival, travel time, and probabilities [14] R. Diaz, Characteristics of bus rapid transit for decision-making.
of system output, which can help to develop investment Federal Transit Administration, 2004.
possibilities. [15] S. S. H. R. P. (US), C. Systematics, and H. S. C. Group, Performance
measurement framework for highway capacity decision making. Trans-
For future work the authors intend to study other metrics, portation Research Board of the National Academies, 2009.
such as capacity, availability, and system flow. Additionally, the [16] D. Roberts, P. Scrimgeour, D. Freeman, B. Jungwirth, C. Norris,
intention is to introduce costs into the system, thereby attempt- S. Rathwell, D. Reage, R. Takagi, and D. Wohlwill, Bus rapid transit
ing to relate the possible implementation of improvements, service design. American Public Transportation Association, 2010.
derived from the model metrics, with best results in terms of [17] M. Chang, G. Darido, E. Kim, D. Schneck, M. Hardy, J. Bunch,
cost effectiveness, offering alternative planning solutions. The M. Baltes, D. Hinebaugh, L. Wnuk, F. Silver et al., “Characteristics
of bus rapid transit for decision-making,” Tech. Rep., 2004.
impact of faults on system performance is also an area that
requires study, reflecting the profound influence that it has on [18] J. Zhao, W. Li, and X. Hu, “Measuring bus service reliability:
An example of bus rapid transit in changzhou,” SUBMISSION OF
the benefit system. MANUSCRIPTS, vol. 17, no. 2, p. 113, 2014.
[19] W. SHAH, S. A. A. SHAH, and W. KUMAR, “Performance evaluation
R EFERENCES of controlled arrival rate system through matrix geometric method using
transient analysis.”
[1] T. Deng and J. D. Nelson, “Bus rapid transit implementation in beijing:
An evaluation of performance and impacts,” Research in Transportation [20] T. Passos, “Acessibilidade é um dos pontos fracos do brt do recife no
Economics, vol. 39, no. 1, pp. 108–113, 2013. norte/sul (accessibility is a weakness of recife brt north / south),” 2015.
[2] C. V. Mandayam and B. Prabhakar, “Traffic congestion: models, costs [21] C. Arias, A. Castro, W. Colombini, P. Custodio, J. Diaz, K. Fjellstrom,
and optimal transport,” in The 2014 ACM international conference on D. Hidalgo, W. Hook, M. King, L. Wei et al., “Manual de brt: Guia
Measurement and modeling of computer systems. ACM, 2014, pp. de planejamento (bus rapid transit manual),” Ministerio das Cidades
553–554. (Brazilian Ministry of Cities), ITDP, Brasilia, DF, Brazil, 2008.

1278

You might also like