Optimization in Railway Scheduling

M.A. Salido1 , M. Abril1 , F. Barber1 , L. Ingolotti1 , A. Lova2 , P. Tormos2

1
Departamento de Sistemas Informáticos y Computación
2
Departamento de Estadı́stica e I.O y Calidad
Universidad Politécnica de Valencia
Email: {msalido,mabril,fbarber,lingolotti}@dsic.upv.es
Email: {allova,ptormos}@eio.upv.es

Keywords: optimization, constraint optimization, railway scheduling problem.

Abstract: Train scheduling has been a significant issue in the railway industry. Over the last few years, numerous ap-
proaches and tools have been developed to aid in the management of railway infrastructure. In this paper, we
describe some techniques, which was developed in a project in collaboration with the Spanish Railway In-
fractructure Manager (ADIF). We formulate train scheduling as constraint optimization problems and present
two filtering techniques for these problem types. These filtering techniques are developed to speed up and
direct the search towards suboptimal solutions in periodic train scheduling problems. The feasibility of our
problem-oriented techniques are confirmed with experimentation using real-life data. The results show that
these techniques enables MIP solvers such as LINGO and ILOG Concert Technology (CPLEX⃝ c
) to terminate
earlier with good solutions.

1 Introduction ish Railway Infractructure Manager (ADIF) (Mom,

2005). The aim of the project is to offer assistance
Railway transportation has played a major role in in the planning of train scheduling to obtain conclu-
the economic development of the last two centuries. sions about the maximum capacity of the network, to
It represented a major improvement in land trans- identify bottlenecks, to provide support in the resolu-
port technology and has obviously introduced impor- tion of incidents, etc. Besides the mathematical pro-
tant changes in the movement of freight and passen- cesses, a high level of interaction with railway experts
gers. Over the last few years, railway traffic has in- is required to be able to take advantage of their expe-
creased considerably, which has created the need to rience.
optimize the use of railway infrastructures. This is,
however, a very difficult task. Nevertheless, numer- In this paper, we propose two problem-oriented fil-
ous approaches and tools have been developed to aid tering techniques for solving periodic train schedul-
in the management of railway infrastructure. These ing. The train scheduling problem has received con-
systems provide advanced graphical interfaces, but siderable attention in the literature: (Szpigel, 1972) is
they still lack of benefits for automatic planning of the first to propose a branch and bound algorithm for
efficient and robust scheduling. Thanks to develop- train scheduling; (Higgins, 1997) define local search,
ments in computer science and advances in the fields tabu search, genetic and hybrid heuristics; (Cai, 1994)
of optimization and intelligent resource management, illustrate a constructive greedy heuristic. Periodic
railway managers can optimize the use of available timetables for railway networks is usually modeled
infrastructures, obtain more robust timetables and ob- by Periodic Event Scheduling Problem (PESP) (Ser-
tain useful conclusions about capacity of their topol- afini, 1989). It is known that the PESP is NP-hard
ogy. (Serafini, 1989). Approaches to solve PESP instances
We describe some results of a long-term collab- cover backtracking strategies in a branch-and-bound
oration between our research group and the Span- context (Serafini, 1989), genetic algorithms (Nachti-
gall, 1996), and some classes of cutting planes (Odijk,
1
This work has been partially supported by the grant 1994). Furthermore, several European companies are
TIN2004-06354-C02- 01 (MEC, Spain - FEDER) and also working on similar systems. These systems in-
GV04B/516 (Generalitat Valenciana, Spain). clude complex stations, rescheduling due to incidents
(Chiu et al., 2002), rail network capacities (Kaas, determines which stations are used and the stop time
1998), etc. These are complex problems for which required at each station for commercial purposes. In
work in network topology and heuristic-dependent order to perform crossing in a section with a one-way
models can offer appropriate solutions. track, one of the trains should wait in a station. This is
The problem formulation is (traditionally) trans- called a technical stop. One of the trains is detoured
lated into a formal mathematical model to be solved from the main track so that the other train can cross
for optimality by means of mixed integer program- or continue.
ming (MIP) techniques. However, in realtime-
environments, with hundred of trains, in different di- 2.2 Problem Statement
rections, along paths of dozens of stations, with con-
straints about departure and arrival times, generate There are three groups of scheduling rules in our rail-
thousands of inequalities and a high number of vari- way system: traffic rules, user requirements rules and
ables take only integer values. As is well known, topological rules. A valid running map must satisfy
this type of model is far more difficult to solve than and optimize the above rules. These scheduling rules
linear programming models. In our framework, the can be modeled using the following constraints:
formal mathematical model is simplified by filtering
techniques in order to speed up the efficiency of well- 1. Traffic rules guarantee crossing and overtaking
known solvers. operations. The main constraints to take into ac-
count are:
• Crossing constraint: Any two trains going in
2 Preliminaries opposite directions must not simultaneously use
the same one-way track.
The crossing of two trains can be performed only
2.1 Terminology on two-way tracks and at stations, where one of
the two trains has been detoured from the main
A running map contains information regarding rail- track. Several crossings are shown in Figure 1.
way topology (stations, tracks, distances between sta-
tions, traffic control features, etc.) and the schedules • Expedition time constraint. There exists a given
of the trains that use this topology (arrival and depar- time to put a detoured train back on the main
ture times of trains at each station, frequency, stops, track and exit from a station.
junctions, crossings, etc,). A sample of a running map • Reception time constraint. There exists a given
is shown in Figure 1, where several train crossings can time to detour a train from the main track so that
be observed. On the left side of Figure 1, the names of crossing or overtaking can be performed.
the stations are presented and the vertical line repre-
2. User Requirements: The main constraints due to
sents the number of tracks between stations (one-way
user requirements are:
or two-way). The objective of our system is to obtain
a correct and optimized running map taking into ac- • Type and Number of trains going in each direc-
count: (i) traffic rules, (ii) user requirements and (iii) tion to be scheduled.
the railway infrastructure topology. • Path of trains: Locations used and Stop time for
A railway network is basically composed of sta- commercial purposes in each direction.
tions and one-way or two-way tracks. A dependency • Scheduling frequency. The frequency require-
can be: Station: is a place for trains to park, stop ments of the departure of trains in both direc-
or pass through. Each station is associated with a tions. This constraint is very restrictive because,
unique station identifier. There are two or more tracks when crossing are performed, trains must wait
in a station where crossings or overtaking can be for a certain time interval at stations. This in-
performed; Halt: ia a place for trains to stop, pass terval must be propagated to all trains going in
through, but not park. Each halt is associated with a the same direction in order to maintain the es-
unique halt identifier. tablished scheduling frequency. The user can
In Figure 1, horizontal dotted lines represent halts, require a fixed frequency, a frequency within
while continuous lines represent stations. On a rail a minimum and maximum interval, or multiple
network, the user needs to schedule the paths of n frequencies.
trains going in one direction and m trains going in the
opposite direction. These trains are of a given type • Departure interval for the departure of the first
and a scheduling frequency is required. trains going in both the up and down directions.
The type of trains to be scheduled determines the • Maximum slack. This is the maximum percent-
time assigned for travel between two locations on the age δ that a train may delay with respect to the
path. The path selected by the user for a train trip minimum journey time.
 Figure 1: A sample of a running map

3. Topological railway infrastructure and type of straints are composed by the parameters defined by
trains to be scheduled give rise other constraints to user interfaces and database accesses.
be taken into account. Some of them are: The formal mathematical model is presented in Ta-
ble 1. Let’s suppose a railway network with r sta-
• Number of tracks in stations (to perform techni-
tions, n trains running in the down direction, and m
cal and/or commercial operations) and the num-INFORM

trains running in the up direction. We assume that

ber of tracks between two locations (one-way or
two connected stations have only one line connect-
two-way). No crossing or overtaking Time
is allowed
ing them. Ti Ak represents that train i arrives at sta-
on a one-way track,
tion k; Ti Dk means that train i departs from station k;
• Time constraints, between each two contiguous T imeik−(k+1) is the journey time of train i to travel
Stations
stations, from station k to k + 1; T Sik and CSik represent
• Added Station time constraints for technical the technical and commercial stop times of train i in
and/or commercial purposes. station k, respectively; and ETi and RTi are the ex-
In accordance with user requirements, the system pedition and reception time of train i, respectively.
should obtain the bestHaltssolution available so that all The main complexity of the problem derives in
the above constraints are satisfied. Several criteria can solving the MIP problem due to the binary (integer)
exist to qualify the optimality of solutions: minimizePaths variables. These integer variables are generated to
duration and/or number of technical stops, minimize manage disjunctive constraints. If we are able to as-
the total time of train trips (span) of the total schedule, sign values to these integer variables, the linearized
giving priority to certain trains, etc. problem can be solved more efficiently. Therefore,
the main goal of our filtering techniques is to find val-
Tracks
ues for these integer variables. This assignment will
2.3 The Formal Mathematical Model be carried out by means of local search and railway
topological knowledge.
Our formal mathematical model can be described as a
constraint optimization problem, where the main ob-
jective function is to minimize the journey time of all
trains. Variables are frequencies, arrival and departure 3 Filtering Techniques
times of trains at stations and binary auxiliary vari-
ables generated for modelling disjunctive constraints. Given the formal mathematical model presented in
Constraints are composed by user requirements, traf- Table 1, the problem turns into a MIP problem, in
fic rules, and topological constraints. These con- which thousands of inequalities have to be satisfied
 Table 1: Formal Mathematical Model of the railway scheduling problem.
∑i=n ∑i=m
(1) Min i=1 (Ti Ar − Ti D1 ) + j=1 (Tj A1 − Tj Dr );
Subject To
/frequency constraint ∀i = 1..n, ∀k = 1..r
(2) Ti+1 Dk − Ti Dk = F requency;
/Time Constrains ∀i = 1..n, ∀k = 1..r
(3.1) Ti Ak+1 − Ti Dk = T imeik−(k+1) ;
(3.2) Tj Ak − Ti Dk+1 = T imeik−(k+1) ;
/Stations Time Constrains ∀i = 1..n, ∀k = 1..r
(4) Ti Dk − Ti Ak − T Sik = CSik ;
/Constrains to limit journey time ∀i = 1..n, ∀j = 1..m
(5.1) Ti Ar − Ti D1 ≤ (1 + 100 δ
) ∗ T imei1−r ;
(5.2) Tj A1 − Tj Dr ≤ (1 + 100 δ
) ∗ T imejr−1 ;
/Crossing Constrains ∀i = 1..n, ∀j = 1..m, ∀k = 1..r
(6.1) Tj Ak − Ti Dk <= 86400 ∗ Yi−j;k−(k+1) ;
(6.2) Ti Ak+1 − Tj Dk+1 <= 86400 ∗ (1 − Yi,j,k );
/Expedition time constrains ∀i = 1..n, ∀j = 1..m, ∀k = 1..r
(7.1) Tj Ak − Ti Dk − 86400 ∗ (Xi,j − Yi,j,k + Yi,j,k+1 − 1) + ETi <= 0;
(7.2) Ti Ak − Tj Dk − 86400 ∗ (Xi,j − Yi,j,k + Yi,j,k+1 − 2) + ETj <= 0;
/Reception time constrains ∀i = 1..n, ∀j = 1..m, ∀k = 1..r
(8.1) Ti Ak − Tj Ak − 86400 ∗ (Xi,j − Yi,j,k + Yi,j,k+1 − 1) + RTi <= 0;
(8.2) Tj Ak − Ti Ak − 86400 ∗ (Xi,j − Yi,j,k + Yi,j,k+1 − 2) + RTj <= 0;
/Binary Constraints
Xi−j ; ∀i = 1..n, ∀j = 1..m
Yi−j;k−(k+1) ; ∀i = 1..n, ∀j = 1..m, ∀k = 1..r

and a high number of variables only take integer val- problem maintains n ∗ m ∗ r integer variables (Y ′ s)
ues. As is well known, this type of model is far more and n ∗ m integer variables (X ′ s). A railway net-
difficult to solve than linear programming models. work with 100 stations and 100 trains going in each
Our filtering techniques work on the binary variables. direction generates 1.01x106 integer variables. This
These variables are grouped into two sets: technique may significantly reduce the problem size
• Variables Y ′ s. A variable Yi,j,k determines the with a reasonable maximum slack (α ≈ 20%) (given
track between station k and station k + 1 in which by railway operator).
train i crosses with train j. If Yi,j,h−1 = 1 for
h ≤ k and Yi,j,p = 0 for p ≥ k then, the crossing Theorem 2. Filtering technique 1 is sound and
between train i and train j is carried out in station complete.
k. Proof. Soundness. Filtering technique 1 is sound
due to the fact that the set of solutions given by Ta-
• Variables X ′ s. A variable Xi,j determines which ble 1 subsumes the set of solutions obtained by filter-
train (i or j) arrives earlier to the crossing station. ing technique 1. This is because it has removed a set
If Xi,j = 0 train i arrives at the crossing station of binary variables and the constraints in which these
first and train j arrives at the same station later. variables are involved.
Completeness. Filtering technique 1 does not re-
3.1 Filtering Technique 1 move any solution. Thus, this technique will find the
same solution as the one obtained by Table 1. Con-
This technique carries out a filtering over the set of straints of type (5) make the set the of removed con-
constraints from the formal mathematical model pre- straints redundant by filtering technique 1. By contra-
sented in Table 1. Many constraints of type (6) (7) and diction, we assume that there is a solution that filter-
(8) can be removed according to their departure times ing technique 1 does not find. Without loss of gener-
and maximum slacks. If a train going in the down di- ality, we assume a maximum slack of 20%. We can
rection arrives at the destination before a train going distinguish two different cases. (1) The lost solution
in the up direction departs, then both trains will not falls into the maximum slack. This os a contradiction
cross each other. Thus, a huge number of constraints because, under this threshold, the restricted problem
and integer variables we can eliminated. The original is the same as Table 1. (2) The lost solution falls out-
side the maximum slack. This solution is not valid
because it does not satisfy constraints (5.1) and (5.2). Table 2: Pseudo-code of filtering technique 2.
Therefore, filtering technique 1 does not lose any so-
lution.
Filtering technique 2
3.2 Filtering technique 2 { /*Limit the stations where two trains can be crossed,
i.e., the number of integer variables*/
Filtering technique 2 is a metaheuristic based on Fil- DeterminePossibleCrossing();
tering technique 1. This technique carries out a LinearSolution=SolveLinearProblem();
guided local search over the binary variables. Once Crossings=DetectCrossings(LinearSolution);
many integer variables have been removed by filter- n=0;
ing technique 1, a new filtering process on the re- while(Not Solution) {
Solution=SearchCrossCombination(window,Crossings);
duced problem can eliminate other integer variables
n++;
by means of a guided local search. Instead of assign- }
ing a random station as a crossing station between if (Solution)
two opposite trains, filtering technique 2 performs a FinalSolution=SolveCrossingOrder(Solution);
linearized execution where the integer variables have }
been transformed into continues ones. Thus, the
crossing between two trains may not be assigned in
stations but on a track between two stations. This will
be the initial point to start the search to find the sta-
tion where the crossing will finally be performed. The this information. With the data acquired from the
search of each crossing between two opposite trains is database, the system generates the formal mathemati-
bounded by 2n + 1 contiguous tracks. This interval cal model.
is composed by n tracks located before the obtained According to the quality of the required solution
crossing and n tracks located after the crossing. In and the problem size, a filtering technique will be ex-
this way, the resultant subproblem can be seen as a ecuted by one of the following ways:
combinatorial problem, where all combinations must 1. Complete: The process is performed taking into ac-
be performed for guarantee the best possible solution. count the entire problem. This decision is carried
If the problem has a solution, filtering technique 2 out when the number of trains and stations is low
studies the arrival order to the crossing station such or the running time is not a important. In this case,
as the objective function is minimized. Otherwise, filtering technique 1 will be selected.
the interval is increased (n + +) and the MIP prob-
2. Incremental: The process performs an incremental
lem is again solved. This technique is useful in any-
coordination of trains. It can be useful in anytime
time environment due to a solution can be found, but
systems, where the number of trains and stations
filtering technique 2 tries to find a better solution in
is not very high. In this case, filtering technique
the remaining time. To this end, each combination is
1 and 2 are appropriate due to the fact that as the
labelled with the solution obtained and the filtering
number of combinations are checked, the quality
technique searches neighbor combination in order to
of the solution is better.
improve the objective function.
Once the problem has been filtered, the optimiza-
tion process will be executed for obtaining an opti-
mal solution of the simplified problem. To this end,
4 GENERAL SYSTEM CPLEX and LINGO are executed for obtaining the
ARCHITECTURE optimal solution.
However, the system can also automatically recom-
The general outline of our system is presented in mend or select the appropriate choice depending on
Figure 2. It shows several steps, some of which re- main parameters and the complexity of the problem.
quire the direct interaction with the human user to If the mathematical model is not feasible, the user
insert requirement parameters, parameterize the con- must modify the parameters, mainly the most restric-
straint solver for optimization, or modify a given tive ones. If the running map is consistent, the graphic
schedule. First of all, the user should require the pa- interface plots the scheduling. Afterwards, the user
rameters of the railway network and the train type can graphically interact with the scheduling to mod-
from the central database (Figure 2). ify the arrival or departure times. Each interaction
This database stores the set of locations, lines, is automatically checked by the constraint checker in
tracks, trains, etc. Normally, this information does order to guarantee the consistency of changes. The
not change, but authorized users may desire to change user can finally print out the scheduling, to obtain re-
 Figure 2: General scheme of our tool.

ports with the arrival and departure times of each train

in each location, or graphically observe the complete
scheduling topology.

4.1 A Constraint-based System for

Automatic Railway Scheduling
Our techniques described here can be extended to
general problems such as: inserting a new train or
several new trains in an already compatible running
map and optimizing running maps.General In this case, new
Scheme
constraints must be taken into account such as clos-
ing constraints, exclusiveness constraints, precedence
Starting Interface

constraints, etc. Figure 4 shows an exampleFiltering

of a high
Optimization
loaded running map with theFormal
proposed system.
Technique 1
Process
Mathematical
Out tool provides severalModelbenefits for automatic Figure 4: Example of a high loaded running map.
Filtering
railway infrastructure scheduling. Technique 2 Running Map
Data Data
• It is flexible
Base and friendly. It can be easily inte-
grated into already existing data-bases and other trains (speeds, starting and stopping times, etc.), fre-
computer-aided tools for railway management. Graphic Interface
quency ranges, initial departure interval times, etc.
• It automatically obtains optimized
Constraint
and well-formed
Plotter
In this section, we compare the performance of
Checker
running maps (timetables). our filtering techniques using some well-known con-
straint optimization problem solvers, CPLEX and
• It can validate complex timetablesINFORM
by automatically LINGO,
Timetabling because they are the most appropriate tools
performing all the consistency checks required for for solving these types of problems.
well-formed timetables. This empirical evaluation was carried out on a real
• It can validate and perform capacity analyses. railway infrastructure that joins two important Span-
• It can reschedule running maps according to inci- ish cities (”La Coruña” and ”Vigo”). The journey be-
dences and delays in on-line, traffic management. tween these two cities is currently divided by 40 de-
pendencies between stations (23) and halts (17).
In our empirical evaluation, each set of instances
was defined by the 3-tuple < n, s, f >, where n was
5 EVALUATION the number of trains in each direction, s the number of
stations/halts and f the frequency. The problems were
The application and performance of this system generated by modifying these parameters. Thus, each
depends on several factors: Railway topology (lo- of the tables shown sets two of the parameters and
cations, distances, tracks, etc.), number and type of varies the other one in order to evaluate the algorithm
 Figure 3: System Interface solving and plotting instance <10,40,90>.

performance when this parameter increases. It must ney time in problems where the number of stations
be taken into account that running time of the form was increased from 10 to 60, and the number of trains
”> xh.” represents that the problem did not finish in and the frequency were set at 10 and 90, respectively:
x hours and the best solution found up to date is pre- < 10, s, 90 >. In this case, only stations were in-
sented in the journey time column. All running times cluded to analyze the behavior of the techniques. It
in Table 3 represent the running times of the filtering can be observed that Filtering technique 2 was bet-
techniques plus the running times of the optimization ter than Filtering technique 1 obtaining optimal solu-
techniques (CPLEX or LINGO). tions for 10 and 20 stations in lower time. Eve for
In Table 3 (a), we present the running time and the 30 stations Filtering technique 2 had better behaviour
journey time in problems where the number of trains than Filtering technique 1 (complete algorithm). It
was increased from 5 to 75, and the number of sta- is important to note the difference between the in-
tions/halts and the frequency were set at 40 and 90, stance < 10, 40, 90 > of the Table 3 (a) and the in-
respectively: < n, 40, 90 >. The results show as the stance < 10, 40, 90 > in Table 3 (b). They repre-
number of trains increased the running time of Filter- sent the same instance; however in Table 3 (b) we
ing technique 1 and 2 was worse. Filtering technique only used stations (no halts), so the number of pos-
1 obtained the optimal solution for 5,10,15 and 20 sible crossing between trains was much larger (more
trains. However for 50 and 75 trains, Filtering tech- integer variables). This item reduced the journey time
nique 1 was aborted in 5 hours while Filtering tech- from 2:20:19 to 2:20:10, using Filtering technique 1
nique 2 finished although with worse solutions. Fig- and from 2:26:04 to 2:23:36, using CPLEX and Filter-
ure 3 shows the system interface executing our Fil- ing technique 2. Nevertheless, the running time also
tering technique 2 with the instance < 10, 40, 90 >. increased from 337” to 2131 in Filtering technique 1
The first window shows the user parameters, the sec- and from 8” to 56” in Filtering technique 2, due to the
ond window presents the best solution obtained at that number of integer variables was much larger.
point, the third window presents data about the best In Table 3 (c), we present the running time and the
solution found, and finally the last window shows the journey time in problems where the frequency was de-
obtained running map. creased from 140 to 60 and the number of trains and
Table 3 (b) shows the running time and the jour- the number of stations were set at 20 and 40, respec-
 Table 3: Running time and journey time in different problem instances.
CPLEX LINGO
(a) < n,40,90> Filtering technique 1 Filtering technique 2 Filtering technique 2
Trains running time journey running time journey running time journey
time time time
5 6” 2:19:48 4” 2:29:33 6” 2:30:54
10 337” 2:20:19 8” 2:22:08 12” 2:31:37
15 601” 2:20:29 12” 2:26:18 19” 2:31:51
20 1065” 2:20:34 16” 2:26:25 25” 2:31:58
50 > 5h. 2:20:43 43” 2:31:09 1098” 2:32:11
75 > 5h. 2:22:04 > 1h. 2:32:14 1590” 2:32:14
(b) <10,s,90>
10 3” 0:25:06 2” 0:25:06 4” 0:25:06
20 303” 1:04:11 5” 1:04:11 8” 1:04:11
30 > 1h. 1:45:38 6” 1:45:08 14” 1:45:38
40 2131” 2:20:10 56” 2:23:36 21” 2:24:36
60 > 3h. 3:33:15 217” 3:39:30 180” 3:40:30
(c) <20,40,f >
140 15” 2:16:19 15” 2:20:18 24” 2:16:19
120 156” 2:16:17 14” 2:16:17 23” 2:18:47
100 > 5h. 2:22:55 15” 2:23:10 28” 2:22:55
90 1065” 2:20:34 15” 2:26:25 28” 2:31:58
75 > 1h. 2:29:18 > 1h. - 25” 2:24:16
60 > 1h. 2:21:23 > 1h. - > 1h. -

tively: < 20, 40, f >. As the frequency decreased, REFERENCES

the process solving become harder. The quality of
the solutions depends mainly of the network topology. Cai, X., G. C. (1994). A fast heuristic for the train schedul-
For this reason, Filtering technique 2 obtained the ing problem. Computers and Operations Research 21,
same solutions than Filtering technique 1 but lower pages 499–510.
running times with frequencies of 140, 120 and 100 Chiu, C., Chou, C., Lee, J., Leung, H., and Leung, Y.
minutes. It can be observed that depending on the (2002). A constraint-based interactive train reschedul-
problem topology, one technique may be better than ing tool. Constraints, 7:167–198.
the others. Therefore, it may be useful for the system Higgins, A., K. E. F. L. (1997). Heuristic techniques for
to automatically select the appropriate technique. single line train scheduling. Journal of Heuristics 3,
pages 43–62.
Kaas, A. (1998). Methods to calculate capacity of railways.
6 CONCLUSIONS Ph. Dissertation.
Mom (2005). An automated decision sup-
We have reported the design and development of port tool for railway traffic management.
two filtering techniques for solving periodic train http://www.dsic.upv.es/users/ia/gps/MOM/.
scheduling, which is a project in collaboration with Nachtigall, L., V. S. (1996). A genetic algorithm approach
the National Network of Spanish Railways (RENFE), to periodic railway synchronization. Computers and
Spain. We have formulated the train scheduling as Operations Research, 23:453463.
constraint optimization problems. Two filtering tech- Odijk, M. (1994). Construction of periodic timetables, part
niques are developed to speed up and direct the search 1: A cutting plane algorithm. TU Delft Technical Re-
towards sub-optimal solutions. The feasibility of our port, pages 94–61.
algorithms are confirmed with experimentation using Serafini, P., U. W. (1989). A mathematical model for peri-
real-life data. These techniques have been inserted odic scheduling problems. Computers and Operations
into the system to solve periodic timetables more effi- Research, 2:550581.
ciently. This system is already integrated and assist to Szpigel, B. (1972). Optimal train scheduling on a single
railway managers in optimizing the use of railway in- track railway. M. Ross, OR ’72, pages 343–351.
frastructures and will also help them in the resolution
of complex scheduling problems. It supposes the ap-
plication of methodologies of Artificial Intelligence in
a problem of great scientific and commercial interest.