OR Spectrum (2012) 34:491–510

DOI 10.1007/s00291-011-0251-6

REGULAR ARTICLE

Line planning in public transportation:

models and methods

Anita Schöbel

Published online: 3 May 2011

© The Author(s) 2011. This article is published with open access at Springerlink.com

Abstract The problem of defining suitable lines in a public transportation system

(bus, railway, tram, or underground) is an important real-world problem that has also
been well researched in theory. Driven by applications, it often lacks a clear description,
but is rather stated in an informal way. This leads to a variety of different published
line planning models. In this paper, we introduce some of the basic line planning
models, identify their characteristics, and review literature on models, mathematical
approaches, and algorithms for line planning. Moreover, we point out related topics
as well as current and future directions of research.

Keywords Line planning · Public transportation · Mathematical programming

1 The line planning problem in public transportation

Given the increasing demand for mobility, an efficient organization of public (passen-
ger) transportation becomes more and more important. This is reflected not only in
practice but also by an increasing number of research papers dealing with the optimi-
zation of public transport. The goal of the optimization process is on the one hand, to
offer a high quality of service for the passengers while, on the other hand, the costs
for setting up and running the transit system should be small.
As noted by many authors (see e.g., Ceder and Wilson 1986; Liebchen and Möhring
2007; Desaulniers and Hickman 2007), the planning process in public transportation
consists of several consecutive planning phases. As shown in Fig. 1, the process starts

A. Schöbel (B)
Institut für Numerische und Angewandte Mathematik,
Georg-August Universität Göttingen, Göttingen, Germany
e-mail: [email protected]

123
492 A. Schöbel

Fig. 1 The planning steps in

public transportation in their infrastructure
classical order

lines and frequencies

timetable

vehicle routes and schedules

crew schedules

with network design, usually followed by line planning, timetabling, and vehicle and
crew scheduling.
In this paper, we review the line planning process. We assume that the infrastructure
is already given and represented as a public transportation network (PTN). In partic-
ular, we assume the stations to be fixed and a set of possible edges to be given. These
can be streets (in bus transportation) or the track system (in rail, tram, or underground
transportation). We are looking for a set of lines (a formal definition will be given
in the next section) in the PTN along which service will be offered. Line planning
hence includes to determine the number and the routes of the lines. It also includes the
determination of the frequencies of the lines, i.e., how often service should be offered.
A line plan together with the frequencies of the lines is called a line concept. The line
planning problem is to find a line concept
• which is feasible in the sense that it can be operated,
• which ensures that public transport is convenient for the passengers, and
• whose costs are small.
We remark that lines together with a rough time frame have been also denoted as
(periodic) service intensions (see Caimi et al. 2009) as the line concept contains the
customer-relevant information about the public transport system.
Stating the line planning problem as above leaves most modeling aspects open. In
particular, there are many different ways to model the costs of a line concept, and
there are even more possibilities to evaluate its quality for the passengers. Moreover,
different constraints (ensuring feasibility) may be considered. This may explain why
there exist so many different models for the line planning problem.
The remainder of the paper is structured as follows. In the next section, we introduce
the basic notation and discuss characteristics of line planning models. We also present
some “basic” and some common line planning models. In Sect. 3, we review literature
on these line planning models including modeling aspects and solution approaches.

123
Line planning in public transportation: models and methods 493

Finally, recent directions of research are sketched in Sect. 4, and suggestions for future
research are identified in Sect. 5.

2 Models for the line planning problem

2.1 Modeling the line planning problem

In this section, we will identify the basic ingredients of the line planning models treated
in the literature. In order to describe them properly, we need some notation.
Notation 1 Let P T N = (V, E) be a public transportation network with its stops or
stations V and its direct connections E.
The PTN describes the underlying street or track network which we assume to be
given and fixed. In this section, we assume that all lines will be operated by homoge-
neous vehicles, i.e., we only consider one single transport mode. This is certainly not
a realistic assumption, but nevertheless makes algorithmically sense since different
modes of transport are often considered one after another. In Sect. 3, we will mention
generalizations of these assumptions made in the literature.
Notation 2 A line l is a path in the public transportation network PTN.
The frequency fl of a line l says how often service is offered along line l within a
given time period I (e.g., an hour, a day).
A line concept (L, f ) is a set of lines L together with their frequencies fl for all l ∈ L.
Note that the frequencies determine how often a line is served within the period I,
but leave it open if this happens in regular intervals (as done in a cyclic timetable) or
in an aperiodic way. It is important to choose the interval I appropriate for the goals
one has in mind: If a special line concept for the morning peak should be determined
then I should refer to the time span of the morning peak; if the line concept should be
the same for all periods of the day then I can be chosen as, e.g., a month.
Also there may be several restrictions on the lines, e.g., it is sometimes required
that lines should be symmetric (i.e., operated in both directions).
As mentioned before, there are two conflicting goals when determining a line con-
cept. On the one hand, costs should be small; on the other hand, the line concept
should be as good as possible for the passengers. Consequently, there are cost- and
passenger-oriented components in all models for the line planning problem. We will
call a model passenger-oriented if its objective is to maximize the quality of the line
concept (costs may be restricted in the constraints). Analogously, a model in which the
goal is to minimize the costs while the constraints ensure a minimum level of quality
for the passengers will be called cost-oriented.
In the following we discuss how cost functions and service quality can be modeled
and evaluated. We will also point out other typical constraints in line planning models.

Modeling the costs

In order to determine the costs of a line concept, data about the expenses for vehi-
cles, drivers, etc. need to be known. Often they are aggregated to some fixed cost

123
494 A. Schöbel

per line and some variable cost which depends on the frequency. The most common
approximation of the costs of a line concept (L, f ) is

c(L, f ) = costl fl . (1)
l∈L

where costl describes the costs for operating one vehicle along line l during the plan-
ning period I.
It depends on the length of l, the time needed for a complete trip along line l, and
on the costs per kilometer and per minute driving, but is independent of the frequency
fl . Many details about the calculation of more realistic costs may be considered with
respect to the specific transportation mode such as, e.g., the number of cars of a train
in railway transportation (see Claessens et al. 1998). We remark that the final costs
depend not only on the lines but also on the vehicle and crew schedules, but these
cannot be taken into account since they are not known in the line planning stage.
In order to keep the costs in a passenger-oriented model small enough c(L, f ) can
be bounded by a budget constraint

(BUD) costl fl ≤ B (2)
l∈L

for some given budget B ≥ 0.

Modeling the quality for the passengers

Also for dealing with the passengers’ data, it is helpful to assume one single transit
mode served by a homogeneous fleet. If this is not the case, the first step suggested by
many authors is the system split procedure (Oltrogge 1994; Bouma and Oltrogge 1994)
which distributes the passengers to the different modes of transport and hence results
in one problem for each transport mode. Note that this cannot always be done, e.g., if
a new rapid transit line within an existing transport system is to be introduced, or if
the different transport modes cannot be separately investigated. Such generalizations
will be pointed out in Sect. 3.
After the system split, we are left with a (single-mode) line planning problem for
each type of transportation. The following two types of data about the passengers may
be considered.

Notation 3 (Wuv )u,v∈V denotes the origin–destination matrix (OD-matrix), i.e.,

Wuv is the number of passengers who want to travel from an origin (station) u ∈ V
to a destination (station) v ∈ V within the planning period I. If Wuv > 0 we say that
(u, v) is an OD-pair.
The traffic load we is the number of passengers traveling along edge e ∈ E within
the planning period I.

In order to evaluate the quality of a system for the passengers, the most common
objective functions are based on the following notation.

123
Line planning in public transportation: models and methods 495

Notation 4 A direct traveler is a passenger who does not have to transfer between
different lines on his way between his origin u and his destination v, u, v ∈ V.
The riding time of a passenger is defined as the time the passenger sits in a bus/train
on his way between his origin and his destination. It neglects the time needed for
transfers.
The traveling time of a passenger consists of the riding time and a penalty (time)
for every transfer.

The exact traveling time can only be calculated if the timetable is known (which is
not the case at the stage of line planning). Using a penalty for each transfer is hence
an approximation of the real transfer time.
Common objective functions are to maximize the number of direct travelers, to
minimize the sum of all riding times of all passengers, or to minimize the sum of all
traveling times of all passengers.
Let us now consider constraints which can be used in a cost-oriented model to
ensure a minimum level of quality for the passengers. The most simple ones are based
on the traffic loads we which are often assumed to be known beforehand. Given a line
concept (L, f ) the capacity constraints

(CAP) capl fl ≥ we for all e ∈ E (3)
l∈L:e∈l

ensure that we passengers can be transported along edge e, where capl denotes the
capacity of a vehicle operating line l, i.e., the number of seats in the bus, tram, under-
ground, or train, respectively. Note that models using (CAP) assume the capacities to
be fixed and known in advance.
In some papers, in particular, when the goal is to maximize the number of direct trav-
elers, a modified capacity constraint (CAP’) is used which requires enough capacity
on every edge for each single line instead of looking at the sum of lines.
An even simpler constraint which guarantees a minimum level of service is to look
at the edge frequencies, i.e., the number of vehicles running along the edges. A line
concept (L, f ) satisfies the lower edge frequency requirements if

(LEF) fl ≥ f emin for all e ∈ E. (4)
l∈L:e∈l

The parameter f emin is often defined as the minimal number of vehicles needed to
transport all passengers: If the capacity cap is the same for all vehicles, and we is an
approximation of the number of passengers traveling along edge e the lower frequency
we
requirements with f emin :=  cap  are equivalent to the capacity constraints (CAP).
If a set of OD-pairs is given, a minimal requirement is that all passengers can travel
between their origins and their destinations, no matter how long the journey takes and
how many transfers are necessary, i.e., every OD-pair should be connected by the lines
in the line concept. This condition is denoted as (CON).
Apart from the different objectives and constraints, there is one more crucial dif-
ference in passenger-oriented line planning models: Up to the beginning of this cen-

123
496 A. Schöbel

tury, mathematical line planning models used a given OD-matrix only to distribute
the passengers on paths in the transportation network before the lines were known.
This first step is known as traffic assignment and generates the traffic loads we by
routing passengers through the PTN along suitable paths. Many procedures for traf-
fic assignment are known. These include shortest path approaches or approaches in
which one assumes that passengers are likely to change to fast transportation modes
(e.g., high-speed trains) as early as possible and leave them as late as possible. The
dependencies between the passengers’ paths is considered in equilibrium models. We
do not review the literature on traffic assignment here but refer to Patriksson (1994),
Carraresi et al. (1995), Peeta and Ziliaskopoulos (2001), Desaulniers and Hickman
(2007), and Babonneau and Vial (2008) and references therein.
However, the real passengers’ weights along every edge strongly depend on the
line concept which is to be designed, leading to a chicken-egg-problem. Hence, new
approaches in which the passengers’ routes are part of the optimization and are not
fixed beforehand, become more and more popular. Such approaches integrate line
planning and traffic assignment. The underlying mathematical model is the change &
go graph developed in Schöbel and Scholl (2006a). It allows to include the route choice
as (individual) shortest path problem for every OD-pair within the optimization.

Notation 5 Given a public transportation network PTN = (V, E) and a set of poten-
tial lines L0 , its corresponding change & go graph consists of a set of nodes
 
(v, l) : v ∈ Vis a station of line l ∈ L0

for every line-station combination and a set of edges

{((v, l1 ), (u, l2 )) : Either (v = u) or (l1 = l2 and (u, v) ∈ E)} ,

i.e., two nodes of the change & go graph are linked if they are consecutive stations
within the same line or if they belong to the same station. The latter case allows to
model transfers between the corresponding lines.
By calculating shortest paths in the change & go graph the individual traveling time
of a passenger traveling through the public transportation network can be evaluated
keeping also track of the number of transfers necessary. Note that different weights
for different transfers are possible, e.g., one can model long transfer times at large
stations.

Technical constraints

Various technical restrictions ensuring the operability of a line concept are also treated
in line planning models. Symmetric to the lower edge frequency requirements (see
(4)) often upper edge frequency requirements are used. For a line concept (L, f ) these
are given as

(UEF) fl ≤ f emax for all e ∈ E (5)
l∈L:e∈l

123
Line planning in public transportation: models and methods 497

where f emax models restrictions such as security headways or noise avoidance. It some-
times is also used to bound the costs of the line concept in passenger-oriented models.
Note that (LEF) and (UEF) have first been introduced by Wegel (1974).
Similar to the edge frequency requirements, also lower/upper node frequency
requirements may be considered. They restrict the number of vehicles stopping at
particular stations using lower bounds f vmin , or upper bounds f vmax , respectively. For
a line concept (L, f ) they are defined as

(LNF) fl ≥ f vmin for all v ∈ V , (6)
l∈L:v∈l

(UNF) fl ≤ f vmax for all v ∈ V . (7)
l∈L:v∈l

Feasibility of lines

Line planning models also differ with respect to the restrictions on the lines which
may be chosen for the line concept. In the literature, two major possibilities to model
feasibility of lines are considered.
The first one assumes that a line pool L0 of potential lines is given. It may have
been collected by the transportation company and consists of what is considered as
“reasonable” lines. All of the lines in the line pool may be established. The goal is
to choose a subset of lines from the line pool, i.e., it is required that L ⊆ L0 . This
approach can be seen as the second phase of a two-step approach: the lines are con-
structed in a first phase, and a line plan is chosen from this set in a second phase.
A distinction into these two steps and heuristics constructing lines for the first phase
have for example been suggested in Ceder and Wilson (1986). Note that most line
planning models use this approach and hence deal with the selection of a set of lines
out of a given line pool L0 . Various models and algorithms exist for constructing a
set of reasonable routes, many of them are based on shortest path procedures, some
of them driven by the passengers’ demand. We refer to Kepaptsoglou and Karlaftis
(2009, p. 498–499) for an overview about such methods.
In the second approach, no predefined lines are given, but the lines are constructed
from scratch within the optimization process. This has often been done heuristically
(see Sect. 3.5) and more recently also in an exact way (Borndörfer et al. 2007). Note
that constructing the lines from scratch does usually not mean that all possible paths
in the PTN may be chosen as lines. Often, there are rules that have to be respected
during the construction, or lines have to satisfy certain criteria regarding their shapes.
The characteristics mentioned in these sections are summarized in two tables.
Table 1 collects the objective functions which are common in line planning models
while Table 2 lists some of the basic constraints.

2.2 Complexity of line planning models

In order to discuss the complexity of line planning models we define the following
basic model:

123
498 A. Schöbel

Table 1 Objective functions in

Cost-V Variable costs of lines are considered See Eq. (1)
line planning models
Cost-F Fixed costs of lines are considered
Cost-FV Fixed and variable costs of lines
are considered
Pass-DT Number of direct travelers Notation 4
Pass-RT Riding time Notation 4
Pass-TT Traveling time Notation 4

Table 2 Basic constraints in

BUD Budget constraint (2)
line planning models
CAP Capacity constraint (3)
CAP’ Capacity constraint for single lines Page 495
LEF Lower edge frequency requirements (4)
UEF Upper edge frequency requirements (5)
LNF Lower node frequency requirements (6)
UNF Upper node frequency requirements (7)
CON Connected path for every OD-pair Page 495

(LP-basic) Given a PTN, a set L0 of potential lines, and lower and upper frequencies
f emin ≤ f emaxfor all e ∈ E, find a line concept (L, f ) with L ⊆ L0 , fl ∈ N0 ∀ l ∈ L,
and f e ≤ l∈L:e∈l fl ≤ f emax for all e ∈ E.
min

Although (LP-basic) looks rather simple, it is already NP-hard as shown in Bussieck

(1998) by exact cover by 3-sets (X3C), even for the special case that f emin = f emax = 1
for all e ∈ E. Since most other models contain (LEF) and (UEF) as constraints, this
result shows NP-hardness for most line planning models.

There are two special cases of (LP-basic) which are easy to solve:
• If upper frequencies are neglected (i.e., f emax = ∞ for all e ∈ E) a feasible solution
exists if every edge with positive f emin is contained in at least one line, and it can be
found easily by adding enough frequency to the lines until all lower edge frequency
requirements are satisfied.
• If all paths of the PTN are allowed as potential lines (LP-basic) is always feasible
and can also be solved in polynomial time by taking each edge as a line, e.g., with
frequency f emin .
These simple cases are sometimes used as starting points for heuristic approaches.
We now add a cost objective function to (LP-basic) and obtain a cost-oriented
model.

Basic cost model (LP-cost) Given a PTN, a set L0 of potential lines, lower and upper
frequencies f emin ≤ f emax for all e ∈ E, and costs costl for each line l ∈ L0 find a line
concept (L,f ) which satisfies the constraints of (LP-basic) and minimizes the costs
c(L, f ) = l∈L costl fl .
In contrast to (LP-basic), (LP-cost) is NP-hard even in the two special cases

123
Line planning in public transportation: models and methods 499

• in which we have no upper frequencies, and

• in the case where all paths of the PTN are allowed as lines.

The first is even true if costl = 1 for all lines and f emin = 1 for all edges as can be
seen by reduction to set covering. Note that (LP-cost) without upper frequencies is a
multi-covering problem and can hence be solved by integer programming approaches
for multi-covering. The latter special case is NP-hard by reduction to Hamiltonian
path.
We remark that by reduction to vertex cover Claessens et al. (1998) showed NP-hard-
ness of (LP-cost) without upper frequencies, but with an additional capacity constraint
(CAP).
We now turn our attention to passenger-oriented models. Many of the early papers
about line planning deal with the direct travelers approach (see Dienst 1978). In this
model, it is assumed that customers use preferable paths (e.g., shortest paths) which
are known beforehand and are fixed for each OD-pair (u, v), u, v ∈ V. The costs in
the direct travelers approach are bounded due to upper edge frequency requirements
(UEF). The problem is NP-hard since (LP-basic) is included as a special case.

(Direct-travelers) Given a PTN, a set L0 of potential lines, lower and upper frequen-
cies f emin ≤ f emax for all e ∈ E, and an OD-matrix Wuv together with fixed paths for
each OD-pair (u, v) for all u, v ∈ V. The goal is to find a line concept (L, f ) which
satisfies the constraints of (LP-basic) and maximizes the number of direct travelers.
More recently, models have been suggested that look not only at direct travelers
but consider the traveling times of all passengers in the system. Such traveling time
models usually allow that passengers are routed freely within the optimization. Costs
are controlled by a budget constraint.

(Traveling-time) Given a PTN, a set L0 of potential lines, costs cl for each line
l ∈ L0 , a budget B, and an OD-matrix Wuv for all u, v ∈ V. The goal is to find a line
concept (L, f ) which satisfies (BUD) and minimizes the sum of traveling times of the
passengers.
Also this problem is NP-hard, even if the PTN is a linear graph and all costs cl are
equal to one; see Schöbel and Scholl (2006a).

3 Literature review

Having discussed the most common models, we will now review literature on line
planning. We will sketch generalizations of the basic models and review solution
approaches. For literature up to the beginning of the 1990s, we refer to the surveys by
Israeli and Ceder (1995) and Chua (1984) for more application-oriented approaches.
Our survey is structured as follows. We start with literature on the classical mod-
els discussing cost-oriented models in Sect. 3.1 and passenger-oriented models in
Sect. 3.2. We then proceed with more recent models, in particular game-theoretic
models in Sect. 3.3 and location-based models for line planning in Sect. 3.4. Sec-
tion 3.5 is devoted to iterative approaches.

123
500 A. Schöbel

3.1 Cost-oriented models

Extensive research on cost-oriented models is presented by Claessens et al. (1998); see

also Zwaneveld (1997), Claessens (1994), and Zwaneveld et al. (1996). The goal of
this work was to determine lines that minimize the operational costs subject to service
constraints and capacity requirements. The (railway) model presented in Claessens
et al. (1998) determines not only the lines and their frequencies, but also the type of
train operating a line and the number of cars for each train. The costs investigated
include fixed costs per car per hour (e.g., cost of capital, fixed maintenance), variable
costs per car per kilometer (as energy costs, cleaning costs, or costs for the ticket col-
lector), and variable costs per train per kilometer (e.g., for the driver and for energy).
These costs may vary for different types of cars and trains. Using decision variables
of type X ltc which are defined to be one, if line l is served by vehicles of type t with c
cars, a nonlinear formulation is presented. In order to linearize the model, its dimen-
tfc
sion is further increased, and variables of type X l are used which further include the
frequency f of the lines. The algorithm suggested is based upon constraint satisfac-
tion and a branch & bound procedure. Computations on real-world data of the Dutch
railway system are reported.
The model has been the basis for many other publications that will be described
in the following: Two different linearizations and a cut & branch algorithm for its
solution have been developed in Bussieck (1998). Numerical results using instances
from the German and the Dutch railway system are reported.
A branch & cut approach (neglecting the upper frequency requirements) based on
the models of Claessens et al. (1998) and Bussieck (1998) is proposed in Goossens
et al. (2004) and Goossens (2004). Practical experiences with the Dutch Railway sys-
tem show that the approach runs in reasonable time for real-world instances. Goossens
(2004) further investigates an extension called multi-line planning problem in which
not all trains need to stop at all stations; see also Goossens et al. (2006). The problem
is modeled as a multi-commodity flow problem with a flow for each type of train.
The numerical results show that its solution is better than the solution obtained from
solving a sequence of single-type line planning problems.
A fast procedure for solving the cost model is proposed in Bussieck et al. (2004).
The authors use a nonlinear formulation of the model of Claessens et al. (1998). They
derive lower bounds for three different linearizations. In their algorithm, they combine
linear and nonlinear programming techniques to develop a (heuristic) variable fixing
procedure based on finding a feasible solution in the nonlinear model followed by a
standard branch & bound approach. Valid inequalities further strengthen their formu-
lation. In their numerical study, they show that their algorithm is able to generate good
solutions within a rather small computation time.
Recently, Torres et al. (2008a) analyzed a cost model on simple network topologies
motivated by the structure of the PTN in Quito. It is a cost-oriented model in which a
sum of fixed costs and variable costs is to be minimized and only lower edge frequency
requirements are present. As generalization, the authors allow that fast lines need not
stop at all stations. The model has been investigated for simple network structures as
corridors and trees. It is shown that the model is NP-hard even in these cases (unless

123
Line planning in public transportation: models and methods 501

the fixed costs are zero, the fleet is homogeneous and only lines containing each arc
forward and backward are considered). Nevertheless, the numerical results indicate
that problems whose underlying PTN has a simple structure can be solved by an IP
solver in reasonable time. Also Torres et al. (2008b) deal with the feeder lines for
the same system in Quito. The model only considers variable costs for the lines and
is again investigated for simple line pools and simple PTNs. A polynomial case is
identified on trees if all lines of the line pool start in the same terminal station. In
Borndörfer et al. (2009), the so-called line connectivity problem is defined which is a
line planning problem in which the only requirement (CON) is that all passengers can
(somehow) travel between their origins and their destinations. The authors point out
that the Steiner Tree Problem is a special case of the line connectivity problem and
hence of many line planning problems.

3.2 Passenger-oriented models

Direct travelers approach

The maximization of the number of direct travelers was the focus of many of the
early research papers. The first formulation of (Direct-traveler) as an integer program
is given in Dienst (1978). The solution method proposed in this work is a branch &
bound approach which builds a line partition by adding lines one after another. As
next node in the branch & bound tree the line with the maximal current direct travelers
is chosen in a greedy manner.
The model in Dienst (1978) assumes infinite capacity of the trains. This assumption
is not needed in the direct travelers approach presented by Bussieck et al. (1996) and
Bussieck (1998). They add a capacity constraint which ensures that all direct travel-
ers can be transported. The resulting models they present are formulated as integer
programs and solved by integer programming techniques:
In Bussieck et al. (1996), the edge frequencies are fixed by setting f emin = f emax
for all e ∈ E. An integer programming formulation is presented which is based on
variables di jl representing the number of direct travelers between i and j that use
line l ∈ L0 . Aggregating these di jk variables over all lines together with relaxa-
tion techniques leads to a model that can be solved by a cutting plane approach.
A slight modification of the model is investigated in Zimmermann et al. (1997). Here,
f emin ≤ f emax is allowed such that the edge frequencies are not fixed in advance.
Moreover, a budget constraint is introduced. The resulting model is formulated as
an integer program and solved by column generation and valid inequalities. Numer-
ical results using data of German, Dutch, and Swiss railways are reported. Many
more details can be found in Bussieck (1998), where the problem of maximizing the
number of direct travelers subject to lower and upper edge frequency requirements
and capacity constraints (CAP’) is called revised direct travelers approach. Prepro-
cessing and constraint generation techniques are suggested, and an extensive polyhe-
dral analysis is provided. Moreover, a software package for solving the problem is
developed.

123
502 A. Schöbel

Minimizing the traveling and the riding time

The first integer programming approach which includes routing of passengers while
minimizing the traveling time was presented in Schöbel and Scholl (2003); see also
Schöbel and Scholl (2006a) and Scholl (2005). The goal is to design the lines in such
a way that the sum of all traveling times of all passengers is minimal, respecting a
budget constraint (BUD). Note that the budget constraint is crucial; otherwise, one
would establish direct lines (or the shortest possibility that can be achieved using the
lines from the line pool) for any passenger. In order to formulate the problem as integer
program, the change & go graph (see Notation 5) is used leading to variables x O,D,a
indicating whether the passengers traveling from origin O to destination D use the arc
a of the change & go graph in their shortest path. The problem is tackled by a Dant-
zig–Wolfe decomposition approach. Different relaxations are compared theoretically
and numerically. For practical instances, the authors report that even the lineariza-
tion is time-consuming to solve. In Schmidt and Schöbel (2010), the complexity of
integrating the routing decisions of the passengers into the line planning problem is
investigated, and it is shown that the resulting problem is NP-hard even in very special
cases. Pseudo-polynomial algorithms are provided for the case of linear networks with
some further assumptions.
Independent of Scholl (2005), Borndörfer and Pfetsch (2006) and Borndörfer et al.
(2007, 2008) present a model in which passenger paths can also be freely routed.
Their objective is to minimize the riding time (neglecting the time for transfers) to
which they add fixed costs and variable costs for the line system. They present two
multi-commodity flow formulations: The first formulation concerns the case in which
all paths are allowed as lines and is the first integer programming model in which both
the passengers’ paths and the paths of the lines are not fixed in advance but determined
within the optimization. Their integer programming formulation is related to a service
network model originally presented in Kim and Barnhart (1997). Flow variables for
the passengers and flow variables for the lines ensure that the lines are constructed
from scratch and the passengers are routed freely through the network.
Their second formulation using a variable for each potential line and for each
potential passenger path is the basis for their branch & price approach. The authors
show that the resulting pricing problem is NP-hard (Borndörfer et al. 2008). Generating
the lines dynamically the problem is solved by a column generation approach which
has been tested using the bus system of the city of Potsdam, Germany. In Borndörfer
and Pfetsch (2006), some special cases of the models in Borndörfer et al. (2007, 2008)
are investigated, namely a model with unsplittable routing (all passengers of the same
OD-pair take the same route) and a model where passengers travel along shortest paths.
The models are solved by heuristics and compared theoretically and numerically.
Another formulation also allowing the passengers to be freely routed was devel-
oped in Nachtigall and Jerosch (2008). Similar to Schöbel and Scholl (2006a), they
consider the traveling times of all passengers including a penalty for transfers. Instead
of routing passengers in a change & go graph, they look for a combination of partial
routes which significantly reduces the number of variables needed making the model
numerically tractable. The resulting variables are of type x O,D,u,v,l indicating if the
passengers traveling from O to D use the line l between the stations u and v. In

123
Line planning in public transportation: models and methods 503

their model, they minimize the traveling time using upper edge frequency and upper
node frequency requirements (UEF, UNF) and a budget constraint (BUD) in order to
bound the costs. Again, column generation is used as solution approach together with
a sophisticated pricing operation. The approach was tested using the bus network of
Berlin, Germany.

Other passenger-oriented approaches

Given upper frequency requirements, Puhl and Stiller (2007) look for a line concept
which maximizes the number of passengers that can be transported, i.e., they inves-
tigate how many passengers can maximally be routed along all possible lines. They
show that their problem results in a path constrained network flow problem and that
it is as hard to approximate as maximum clique. Polynomially solvable special cases
are identified.
Klier and Haase (2008) suggest to use frequency-dependent transfer times as input
data. Using a small example, they show how these may be calculated in a pre-pro-
cessing step in which different line combinations for every OD-pair are evaluated, and
up to five possible paths per OD-pair are generated. These paths are weighted by the
expected number of passengers that are likely to use them. The objective maximizes
the total number of expected passengers.

3.3 Game-theoretic models

In order to obtain delay-resistant line plans, Schöbel and Schwarze (2006b) and
Schwarze (2008) present a game-theoretic approach to line planning. The lines cor-
respond to the players that decide about their frequencies. Their individual benefit
functions include the delay-resistance of their lines which does depend on all lines
using the same infrastructure and hence on the decision of the other players. An equi-
librium ensures that the frequencies are equally distributed over the network, and
hence, delays due to capacity conflicts are less likely.
Another game-theoretic approach has been suggested in Kontogiannis and Zaro-
liagis (2008a,b). Similar to Schöbel and Schwarze (2006b), a potentially large number
of line operators is given which have their lines fixed and try to maximize their per-
sonal benefits. In contrast to the former model, Kontogiannis and Zaroliagis (2008a)
introduce a network operator whose duty is the management of the infrastructure. The
goal of the network operator is to achieve a social optimum by maximizing the sum of
benefits of the operators (which he does not know). The authors show how this can be
accomplished in certain situations by suggesting a pricing scheme for the usage of the
shared infrastructure which is robust with respect to the different (uncertain) benefit
functions of the single operators.

3.4 Location-based models

In location theory [see Hamacher and Drezner (2001) for an introduction and over-
view about the field], researchers consider the following problem: Given a network,

123
504 A. Schöbel

where to locate a path in the network such that the distances from the path to a given
set of demand points is minimized? This problem can be interpreted as locating a
rapid-transit line in a given transportation network. The classical objective in loca-
tion theory is to minimize the access times to the new line (location of a median
path). However, recently, also OD-based objective functions were investigated, and
applications in transportation planning are described. In these studies, the new path is
interpreted as a high speed line, and the objective is to locate it in such a way that the
distances for the OD-pairs (which may use the new high-speed line if it is beneficial
for them) are minimal. The papers mentioned in this section follow this approach
and consider the planning of one high speed line in an existing transportation net-
work.
A passenger-oriented approach called trip coverage model has been proposed by
Laporte et al. (2005, 2007). The goal is to design lines which are able to com-
pete with the private mode (e.g., using private cars or bicycles). The objective is to
maximize the number of passengers using public transportation. The model hence
incorporates not only an assignment but also a modal split procedure within the
line planning model. The authors present integer programming formulations and a
case study. Locating a metro line in Seville is considered in Laporte et al. (2009);
metaheuristics for solving the problem are suggested in Martínez et al. (2005). A
solution approach using Bender’s decomposition is proposed in Marin and Jaramillo
(2009).
A systematic analysis of minimizing the traveling times for a given set of OD-pairs
by fastening some of the edges in an existing network has been done by Schmidt
and Schöbel (2009). In their model, they allow all possible paths as lines and use
a budget constraint which fixes the total length of the new high-speed lines but
not their number. It turns out that planning a single high-speed line in a tree net-
work or in a simple circle is polynomially solvable; all other variants are NP-hard.
The paper also investigates the location of other structures, e.g., of a high-speed
tree.

3.5 Heuristics for various line planning problems

So far we discussed publications that all started with defining a model for line planning
and then presented approaches for its solution. However, also the other way round is
followed in the literature: driven by applications, and often without specifying a clear
model, many heuristics are proposed. They aim at constructing a line concept taking
into account various objective functions greedily in each of their steps. We mention
some of the ideas used.
Concerning constructive procedures, the basic idea can already be found in Lampkin
and Saalmans (1967). Starting from a simple set of lines that does not cover all stops
in the network, uncovered stops are inserted sequentially until a connected network
is obtained. In every step, the costs of the insertion should be small. Another idea is
to start with a line plan containing every edge as a single line and to combine these
lines. In Wegel (1974), this is done maximizing the number of direct travelers in every
step.

123
Line planning in public transportation: models and methods 505

Silman et al. (1974) and Dubois et al. (1979) propose the skeleton method in which
they choose the endpoints of the lines and a few intermediate stops which are then
joined by shortest paths. Depending on the actual goal, the shortest paths can be chosen
with respect to minimal length or minimal traveling time.
The algorithm of Sonntag (1977, 1979) [similar to the one of Patz (1925)] starts
with a line plan which contains a line for each OD pair. Lines are then iteratively
eliminated, while new lines may be constructed by joining parts of lines. Passengers
are routed to other paths in every iteration keeping track of their traveling times.
Mandl (1980) directly starts with a feasible set of lines which is heuristically
improved by a neighborhood search allowing to exchange parts of the lines or to
add or remove stations from lines. Starting from an empty set of lines Simonis (1981)
iteratively constructs a line concept in which lines are contained in the set of shortest
paths for a given set of OD-pairs and maximizes the number of direct travelers in each
step. Pape et al. (1995) propose the dual set method. They first identify core lines
containing a large number of travelers which are then combined to a partial line plan.
Other lines are then added to make sure that all edges are covered; in each step, the
number of direct travelers is maximized. Quak (2003) further refines this procedure
by adding a timetabling step behind. He presents a good overview about constructive
heuristics and a case study.
A related area of research concerns literature on the transit route network design
problem (TRNDP) which is defined as a “process which implies clear and consis-
tent techniques for designing a public transportation network” (see Kepaptsoglou and
Karlaftis 2009), i.e., the focus is on defining an approach rather than a model. TRNDP
literature deals with the definition of the routes of the vehicles (which can be later
seen as lines) and their frequencies, but may also comprise other planning stages such
as network design aspects, timetabling, vehicle scheduling, or even the definition of
fares in the tariff system. Nevertheless, the basic restrictions considered in literature
on the TRNDP are similar to restrictions considered in line planning [i.e., (LEF) and
(UEF) are often mentioned], but in TRNDP the constraints considered are usually
much more detailed. They take into account, e.g., load factors, restrictions on the
shape, directness, maximum length and number of routes and even legal requirements
such as bus transit industry guidelines; see Zhao and Zeng (2006) for an overview
on different types of constraints. The goals of TRNDP procedures are cost-oriented,
passenger-oriented, or a combination of both (often called welfare). Also other objec-
tives have been considered such as capacity maximization, minimization of the fleet
size, or energy conservation. It is worth mentioning that there are also approaches fol-
lowing multi-objective goals (see e.g., Baaj and Mahmassani 1991; Israeli and Ceder
1995; Bielli et al. 2002; Chakroborty 2003; Fan and Machemehl 2006; Mauttone and
Urquharta 2009). The procedures proposed for finding a solution to the TRNDP often
stem from real-world (bus) applications and contain similar elements as the line plan-
ning heuristics, e.g., route generation algorithms where in every step new vertices
are inserted (e.g., Baaj and Mahmassani 1991; Mauttone and Urquharta 2009), local
improvements, genetic algorithms, and other metaheuristics. We refer to Kepaptsoglou
and Karlaftis (2009) for a recent and excellent survey on the topic including a list of
more than 60 papers on the TRNDP.

123
506 A. Schöbel

4 Ongoing research in line planning

Robustness in line planning

Nowadays, robustness is more and more an issue in optimization (see Bertsimas and
Sim 2004; Ben-Tal et al. 2009). Given an optimization problem, the classical con-
cepts of strict robustness ask for a solution which is feasible for all possible changes
in the input data (sometimes also called scenarios). It turns out that hedging against
all uncertainties is often much too conservative such that more applicable robustness
concepts are needed for many real-world applications. In particular, for optimization
of public transportation, new concepts to analyze robustness of a plan and to develop
robust plans have been proposed; see the concepts of recovery robustness (Liebchen
et al. 2009; Erera et al. 2009; Cicerone et al. 2009) and of light robustness (Fischetti
and Monaci 2009; Schöbel 2010).
A line concept may be called robust if it is still good in case of changes of the input
data. There are two different types of uncertainties that are to be considered:
• Disturbances during operation. Sometimes a line concept has to be updated during
operation. This may happen if a special event takes place or if a track is not available
due to maintenance or accidents. Scenarios to be taken into account hence only dif-
fer in a few parameters from the expected values of the input parameters, but these
may change drastically.
• Approximation of the input parameters. As in many optimization problems, the
input parameters may be inaccurate. This concerns costs or traveling times, as well
as the information about the passengers. Hence, most of the input parameters are
uncertain, but it can be assumed that each single uncertainty is rather small.
In the game-theoretic approaches described in Sect. 3.3, robust line plans are gen-
erated with respect to delays or with respect to the benefit functions of the operators.
More about robustness in public transportation can be found in the publications of the
European project ARRIVAL and in Ahuja et al. (2009).

Integration with other planning phases

Another issue is integrated model in which the line planning phase is solved simulta-
neously with some other phases of the planning process (depicted in Fig. 1). Due to the
complexity of the integrated problem, exact solution procedures cannot be expected.
We mention some papers containing aspects of integration.
The earliest planning phase is the network design process which includes plan-
ning the stations. Integration of line planning and the location of the stations has
been treated in Chapter 6 of Goossens (2004) and also in some of the location-ori-
ented approaches. Line planning, while respecting network design issues and platform
assignment, is treated in Carey (1994). The integration of line planning with timet-
abling and network design is analyzed in Barber et al. (2008). Integration of line
planning and vehicle scheduling has been investigated in Claessens et al. (1998). The
three planning steps line planning, timetabling, and vehicle scheduling have been
considered in Israeli and Ceder (1995) and Quak (2003), while Liebchen (2008) even

123
Line planning in public transportation: models and methods 507

investigates the integration of the four planning steps line planning, timetabling, vehi-
cle- and crew-scheduling. An integrated model and a heuristic changing the order of
the classical planning steps has been proposed in Michaelis and Schöbel (2009). On the
other hand, line planning has been also considered within timetabling; see Liebchen
and Möhring (2007) and Lindner (2000).

5 Conclusion and directions of future research

In this paper, we reviewed existing literature on line planning problems. We discussed

different modeling aspects of line planning and identified the basic characteristics
of the various existing models making it easier to compare different algorithms and
approaches.
As already mentioned, robustness of line planning models and their integration with
other planning phases are challenging fields of current and future research. In order to
evaluate aspects of integration, we are currently developing a simulation tool called
LinTim (Goerigk et al. 2011). Based on simulated but close-to-real-world data, it can
be used to analyze line planning, timetabling, vehicle scheduling, and delay manage-
ment in an integrated way and to answer questions concerning the delay-resistance of
lines or the effect of line planning on timetabling.
We already pointed out that the two main objectives of line planning contradict
each other: A line concept which is convenient for the passengers is costly, and a line
concept which uses a small budget usually does not offer the service that customers
would like to have. It could hence be a promising task to combine these two objectives
within a bicriteria model and to identify Pareto solutions of such line planning models.
Finally, the line planning models have to go further to practice to make them appli-
cable to the real-world planning process. This can only be achieved by discussing and
adapting the current line planning models keeping in mind (at least) the following
points: For real-world applications, it would be beneficial to be able to determine the
capacities of the vehicles instead of assuming them to be fixed. This can be done
analogously to the extended cost model in Claessens et al. (1998) and should also be
included in other models. Also it needs to be further considered that a good public
transportation system may generate new demand [a first investigation about this issue
can be found in Laporte et al. (2005)]. A more philosophical questions is if peak and
off-peak demand should be treated by different line concepts or if it is appropriate to
use the same line concept all over the day.

Open Access This article is distributed under the terms of the Creative Commons Attribution Noncom-
mercial License which permits any noncommercial use, distribution, and reproduction in any medium,
provided the original author(s) and source are credited.

References

Ahuja RK, Möhring RH, Zaroliagis CD (eds) (2009) Robust and online large-scale optimization. Lecture
Note on Computer Science, vol 5868. Springer, Berlin
ARRIVAL. Future and emerging technologies unit of EC (IST priority, 6th FP), under contract no. FP6-
021235-2. http://arrival.cti.gr

123
508 A. Schöbel

Baaj MH, Mahmassani H (1991) An AI-based approach for transit route system planning and design. J Adv
Transp 25(2):187–210
Babonneau F, Vial J-P (2008) Test instances for the traffic assignment problem. Technical report, Ordecsys
Barber F, Ingolotti L, Lova A, Marin A, Mesa J, Ortega F, Perea F, Tormos P (2008) Integrating timetabling,
network and line design. Technical report, ARRIVAL project
Ben-Tal A, El Ghaoui L, Nemirovski A (2009) Robust Optimization. Princeton University Press, Princeton
Bertsimas D, Sim M (2004) The price of robustness. Oper Res 52(1):35–53
Bielli M, Caramia M, Caratenuto P (2002) Genetic algorithms in bus network optimization. Transp Res
Circ 10(1):19–34
Borndörfer R, Pfetsch ME (2006) Routing in line planning for public transportation. In: Operations research
proceedings 2005, pp 405–410. Springer, Berlin
Borndörfer R, Grötschel M, Pfetsch ME (2007) A column generation approach to line planning in public
transport. Transp Sci 41:123–132
Borndörfer R, Grötschel M, Pfetsch ME (2008) Models for line planning in public transport. In: Computer-
aided scheduling of public transport (CASPT). Lecture notes in economics and mathematical systems,
vol 600, pp 363–378
Borndörfer R, Neumann M, Pfetsch ME (2009) The line connectivity problem. In: Operations research
proceedings 2008, pp 557–562. Springer, Berlin
Bouma A, Oltrogge C (1994) Linienplanung und simulation für öffentliche verkehswege in praxis und
theorie. Eisenbahntechnische Rundschau 43(6):369–378 (in German)
Bussieck MR (1998) Optimal lines in public transport. PhD thesis, Technische Universität Braunschweig
Bussieck MR, Kreuzer P, Zimmermann UT (1996) Optimal lines for railway systems. Eur J Oper Res
96(1):54–63
Bussieck MR, Lindner T, Lübbecke ME (2004) A fast algorithm for near cost optimal line plans. Math
Methods Oper Res 59(3):205–220
Caimi G, Laumanns M, Schüpbach K, Wörner S, Fuchsberger M (2009) The periodic service intention as a
conceptual frame for generating timetables with partial periodicity. In: Proceedings of the international
seminar on railway operations reserach (ISROR)
Carey M (1994) A model and strategy for train pathing with choice of lines, platforms and routes. Transp
Res B 28(5):333–353
Carraresi P, Malucelli F, Pallottino S (1995) On the regional mass transit assignment problem. In:
Sciomachen A (ed) Optimization in industry. 3. Mathematical programming and modeling techniques
in practice, pp 19–34. Wiley, New York
Ceder A, Wilson NHM (1986) Bus network design. Transp Res B 20(4):331–344
Chakroborty P (2003) Genetic algorithms for optimal urban transit network design. Comput Aided Civ
Infrastruct Eng 18(3):184–200
Chua TA (1984) The planning of urban bus routes and frequencies: a survey. Transportation 12(2):147–172
Cicerone S, D’Angelo G, Di Stefano G, Frigioni D, Navarra A, Schachtebeck M, Schöbel A (2009) Recov-
erable robustness in shunting and timetabling. In: Robust and online large-scale optimization. Lecture
notes in computer science, vol 5868, pp 28–60. Springer, Berlin
Claessens MT (1994) De kost-lijnvoering. Master’s thesis, University of Amsterdam, Amsterdam (in Dutch)
Claessens MT, van Dijk NM, Zwaneveld PJ (1998) Cost optimal allocation of rail passenger lines. Eur J
Oper Res 110:474–489
Desaulniers G, Hickman M (2007) Public transit. In: Laporte G, Barnhart C (eds) Transportation. Handbooks
in operations reserch and management science, vol 14, pp 69–127. Elsevier, Amsterdam
Dienst H (1978) Linienplanung im spurgeführten Personenverkehr mit Hilfe eines heuristischen Verfahrens.
PhD thesis, Technische Universität Braunschweig (in German)
Dubois D, Bell G, Llibre M (1979) A set of methods in transportation network synthesis and analysis.
J Oper Res Soc 30(8):797–808
Erera AL, Morales JC, Svalesbergh M (2009) Robust optimization for empty repositioning problems. Oper
Res 57(2):468–483
Fan W, Machemehl RB (2006) Optimal transit route network design problem with variable transit demand:
genetic algorithm approach. J Transp Eng 132:40–51
Fischetti M, Monaci M (2009) Light robustness. In: Ahuja RK, Möhring RH, Zaroliagis CD (eds) Robust
and online large-scale optimization. Lecture note on computer science, vol 5868, pp 61–84. Springer,
Berlin
Goerigk M, Schachtebeck M, Schöbel A (2011) LinTim—integrated optimization in public transportation.
http://lintim.math.uni-goettingen.de/

123
Line planning in public transportation: models and methods 509

Goossens J (2004) Models and algorithms for railway line planning problems. PhD thesis, University of
Maastricht, Maastricht
Goossens J, van Hoesel CPM, Kroon LG (2004) A branch and cut approach for solving line planning
problems. Transp Sci 38:379–393
Goossens J, van Hoesel CPM, Kroon LG (2006) On solving multi-type railway line planning problems.
Eur J Oper Res 168(2):403–424
Hamacher HW, Drezner Z (eds) (2001) Location theory—applications and theory. Springer, Berlin
Israeli Y, Ceder A (1995) Transit route design using scheduling and multiobjective programming tech-
niques. In: Computer-aided scheduling of public transport (CASPT). Lecture notes in economics and
mathematical systems, vol 430, pp 56–75. Springer, Berlin
Kepaptsoglou K, Karlaftis M (2009) Transit route network design problem: Review. J Transp Eng
135(8):491–505
Kim D, Barnhart C (1997) Transportation service network design: models and algorithms. In: Computer-
aided scheduling of public transport (CASPT). Lecture notes in economics and mathematical systems,
vol 471, pp 259–283. Springer, Berlin
Klier MJ, Haase K (2008) Line optimization in public transport systems. In: Operations research proceedings
2007, pp 473–478. Springer, Berlin
Kontogiannis S, Zaroliagis C (2008a) Robust line planning through elasticity of frequencies. Technical
report, ARRIVAL project
Kontogiannis S, Zaroliagis C (2008b) Robust line planning under unknown incentives and elasticity of fre-
quencies. In: Fischetti M, Widmayer P (eds) ATMOS 2008—8th workshop on algorithmic approaches
for transportation modeling, optimization, and systems, Dagstuhl, Germany
Lampkin W, Saalmans PD (1967) The design of routes, service frequencies and schedules for a municipal
bus undertaking: a case study. Oper Res Q 18:375–397
Laporte G, Mesa JA, Ortega FA (2005) Maximizing trip coverage in the location of a single rapid transit
alignment. Ann Oper Res 136:49–63
Laporte G, Marín A, Mesa JA, Ortega FA (2007) An integrated methodology for rapid transit net-
work design. In: Algorithmic methods for railway optimization. Lecture notes in computer science,
vol 4359, pp 187–199. Springer, Berlin
Laporte G, Mesa JA, Ortega F, Pozo M (2009) Locating a metro line in a historical city centre: application
to sevilla. Technical report, ARRIVAL Project
Liebchen C (2008) Linien-, Fahrplan-, Umlauf- und Dienstplanoptimierung: Wie weit können diese bereits
integriert werden? In: Friedrich M (ed) Heureka’08—Optimierung in Transport und Verkehr, Tag-
ungsbericht. FGSV Verlag (in German)
Liebchen C, Möhring R (2007) The modeling power of the periodic event scheduling problem: railway
timetables—and beyond. In: Algorithmic methods for railway optimization. Lecture notes in com-
puter science, vol 4359, pp 3–40. Springer, Berlin
Liebchen C, Lüebbecke M, Möhring RH, Stiller S (2009) The concept of recoverable rrobustness, linear
programming recoery, and railway applications. In: Ahuja RK, Möhring RH, Zaroliagis CD (eds)
Robust and online large-scale optimization. Lecture notes in computer science, vol 5868. Springer,
Berlin
Lindner T (2000) Train schedule optimization in public rail transport. PhD thesis, Technische Universität
Braunschweig
Mandl CE (1980) Evaluation and optimization of urban public transportation networks. Eur J Oper Res
5:396–404
Marin ÁG, Jaramillo P (2009) Urban rapid transit network design: accelerated benders decomposition. Ann
Oper Res 169(1):35–53
Martínez FJ, Melián B, Garzón A, Mesa JA, Moreno J, Ortega FA (2005) Grasp metaheuristic for the
rapid transit network design. In: Actas del IV Congreso Español sobre Metaheurísticos (MAEB),
pp 601–608 (in Spanish)
Mauttone A, Urquharta ME (2009) A route set construction algorithm for the transit network design prob-
lem. Comput Oper Res 36(8):2440–2449
Michaelis M, Schöbel A (2009) Integrating line planning, timetabling, and vehicle scheduling: a customer-
oriented approach. Public Transp 1(3):211–232
Nachtigall K, Jerosch K (2008) Simultaneous network line planning and traffic assignment. In: Fischetti M,
Widmayer P (eds) ATMOS 2008—8th workshop on algorithmic approaches for transportation mod-
eling, optimization, and systems, Dagstuhl, Germany. http://drops.dagstuhl.de/opus/volltexte/2008/
1589

123
510 A. Schöbel

Oltrogge C (1994) Linienplanung für mehrstufige Bedienungssysteme im öffentlichen Personenverkehr.

PhD thesis, TU Braunschweig (in German)
Pape U, Reinecke Y-S, Reinecke E (1995) Line network planning. In: Computer-aided scheduling of public
transport. Lecture notes in economics and mathematical systems, vol 430
Patriksson M (1994) The traffic assignment problem: models and methods. Topics in transportation, VSP
Patz A (1925) Die richtige Auswahl von Verkehrslinien bei großen Strassenbahnnetzen. Verkehrstechnik,
50/51 (in German)
Peeta S, Ziliaskopoulos A (2001) Foundations of dynamic traffic assignment: the past, the present and the
future. Netw Sp Econ 1:233–265
Puhl C, Stiller S (2007) The maximum capacity of a line plan is inapproximable. Technical report, ARRIVAL
Project
Quak CB (2003) Bus line planning. Master’s thesis, TU Delft
Schmidt M, Schöbel A (2009) Location of speed-up networks. Technical report, Institut für Numerische
und Angewandte Mathematik, Georg-August Universität Göttingen (submitted)
Schmidt M, Schöbel A (2010) The complexity of integrating routing decisions in public transportation
models. In: Erlebach T, Lübbecke M (eds) Proceedings of ATMOS10, Dagstuhl, Germany. Open-
Access series in informatics (OASIcs), vol 14, pp 156–169, Schloss Dagstuhl–Leibniz-Zentrum fuer
Informatik
Scholl S (2005) Customer-oriented line planning. PhD thesis, Technische Universität Kaiserslautern
Schöbel A (2010) Light robustness and the trade-off between robustness and nominal quality. Technical
report, Preprint-Reihe, Institut für Numerische und Angewandte Mathematik, Georg-August Univer-
sität Göttingen, Nr. 2010.08 (submitted)
Schöbel A, Scholl S (2003) Planung von Linien mit minimalen Umsteigevorgängen. In: Mattfeld D (ed)
Proceedings of the GOR-workshop on “Optimierung im öffentlichen Nahverkehr”, pp 69–89. http://
server3.winforms.phil.tu-bs.de/gor/tagung34/
Schöbel A, Scholl S (2006a) Line planning with minimal travel time. In: 5th workshop on algorithmic
methods and models for optimization of railways, number 06901, Dagstuhl seminar proceedings
Schöbel A, Schwarze S (2006b) A game-theoretic approach to line planning. In: 6th workshop on algorith-
mic methods and models for optimization of railways, number 06002, Dagstuhl Seminar proceedings
Schwarze S (2008) Path player games: analysis and applications. Springer, Berlin
Silman LA, Brazily Z, Passy U (1974) Planning the route system for urban busses. Comput Oper Res
1:201–211
Simonis C (1981) Optimierung von Omnibuslinien. Technical report, Berichte Stadt-Region-Land, Institut
für Stadtbauwesen, RWTH Aachen (in German)
Sonntag H (1977) Linienplanung im öffentlichen Personennahverkehr. PhD thesis, TU Berlin (in German)
Sonntag H (1979) Ein heuristisches Verfahren zum Entwurf nachfrageorientierter Linienführung im öffent-
lichen Personenverkehr. Z Oper Res 23 (in German)
Torres LM, Torres R, Borndörfer R, Pfetsch ME (2008a) Line planning on paths and tree networks
with applications to the quito trolebús system. In: Fischetti M, Widmayer P (eds) ATMOS 2008—
8th workshop on algorithmic approaches for transportation modeling, optimization, and systems,
Dagstuhl, Germany. http://drops.dagstuhl.de/opus/volltexte/2008/1583
Torres LM, Torres R, Borndörfer R, Pfetsch ME (2008b) On line planning problem in tree networks.
ZIB-Report 08-52
Wegel H (1974) Fahrplangestaltung für taktbetriebene Nahverkehrsnetze. PhD thesis, TU Braunschweig
(in German)
Zhao F, Zeng X (2006) Optimization of transit network layout and headway with a combined genetic
algorithm and simulated annealing method. Eng Optim 38(6):701–722
Zimmermann UT, Bussieck MR, Krista M, Wiegand K-D (1997) Linienoptimierung—modellierung und
praktischer einsatz. In: Hoffmann K-H, Jaeger W, Lohmann T, Schunck H (eds) Mathematik—
Schlüsseltechnologie fuer die Zukunft, pp 595–607 (in German)
Zwaneveld PJ (1997) Railway planning—routing of trains and allocation of passenger lines. PhD thesis,
School of Management, Rotterdam
Zwaneveld PJ, Claessens MT, van Dijk NM (1996) A new method to determine the cost optimal allocation
of passenger lines. In: Defence or attack: proceedings of 2nd TRAIL PhD congress 1996, Part 2,
Delft/Rotterdam. TRAIL Research School

123