Sustainable Global Service Intention as objective for Controlling Railway Network Operations in

Real Time
R. Wst, F.Laube, S.Roos, G.Caimi*
Swiss Federal Railways, Infrastructure, CH 3000 Bern 65, (Switzerland)
*ETH Zurich, Institute for Operations Research, CH 8092 Zrich, (Switzerland)
Abstract
Public transport in Switzerland is based on a regular cyclic timetable connecting long distance and
regional train services with local bus services and tourist ship lines. Thanks to this integrated cyclic
timetable it is possible to offer connections on a point to point network at reasonably high frequencies.
The commercial idea behind this offer can be called global service intention (GSI). In case of operational
disturbances it is a significant challenge for the Swiss Federal Railways (SBB) to find the best substitution
which is available for a delayed or disrupted train service in the given case. The criteria for a good or
even the best substitution must attempt to minimise inconvenience caused to patrons and hence can be
defined in terms of the total delay time of all the passengers concerned by the disturbance.
Some years ago, SBB (together with the ETH Zurich) started to investigate how the GSI can be
decomposed into smaller manageable subunits, the local service intentions (LSI) which are defined for
areas around stations with saturated traffic density. In addition, methods have been developed enabling
LSIs to be automatically converted into local timetables and production plans.
On the other hand a new planning concept is currently being introduced by SBB. This concept is based
on discrete time slots within which trains can enter, exit or pass through saturated station areas. These
time slots are called Pulses and have a temporal extent of approximately the minimum headway in the
entry or exit corridor of the station area. Different trains are able to move either at the same time on non
conflicting routes or within different time slots on conflicting routes.
Due to this discretisation of the scheduling problem into rather coarse but intelligent temporal and spatial
units, the solution space is considerably reduced compared to a continuous scheduling problem. As a
consequence, semi-automatic or even manual rescheduling of arrival, departure and travel times and
simultaneously associating conflict free tracks in real time are possible in practice.
In 2006 SBB has started to implement a feasibility test of the core component of the new concept in
Lucerne. The core component consists of the real time rescheduling based on the monitoring of train
travel and departure times, associating new time slots in case of exceeded tolerances and
communicating the new time slots and new target departure and travel times to the train drivers, train
guards and operators. Thanks to the strict time control the dispersion in departure and travel times can be
significantly reduced (up to 80%). Operational reserves become much more transparent and manageable.
This is an important requirement for increasing capacity of saturated station areas and optimising the
local service intention in case of operational irregularities.
In a next step, several saturated station areas are controlled together. This will allow SBB to optimise
dispatching criteria in terms of the global service intention.
The impact for SBB will be to change the key performance indicator from the average punctuality of trains
to minimum inconvenience caused to patrons measured, for example, as the average total trip delay.
1

Background

1.1 The Concept of Bahn 2000

In Switzerland as in most developed countries, rail transportations main competitor is the car as private
transport plays a key role in mobility. To be competitive, public transport has to provide a maximum of
benefits similar to private traffic. Main arguments for mode choice are travel time from door to door
including convenience and transport availability.
Public transportation as a mass transit mode has distinct characteristics to be taken into account. A
maximum use of trains and buses has to be generated by attributing demand flows to main transport
corridors. This is achieved either by a high demand density along the lines or a hinterland well-linked with
the trunk route system.

Travel times using public transport play a decisive role and are comprised of different components:
accessing a station, waiting, riding, eventual interchanging and transferring from the final station to the
destination. Therefore, successful public transport has to provide a dedicated feeder system to and from
stations, punctual and reliable services together with convenient and quick connections and an effective
customer information system.
In order to approach these aims all together, the passenger transport chain in particular required a
complete redesign. In a first step, Switzerlands successful Rail2000-concept had been introduced
stepwise over 25 years before it was completed in 2005 (For further details see [4]). The design followed
a clear strategy throughout the entire supply chain. The resulting process chain is summarised in
Figure 1.

Demand

Service Intention

Timetable

Production Plan

Figure 1: Design process Rail 2000

1.2 The Supply Chain of Rail 2000

A service pattern that gives an adequate choice of travel opportunities serving a maximum of existing and
potential customers must be offered in order to satisfy demand. High quality public transport combines
different service types (e.g. stopping trains, expresses, etc) in a clever manner. Keep it simple for the
customer is the key of any successful service concept:
Regular stopping patterns and headways as well as connecting with relevant routes in appropriate hubs
of the network provide convenient, quick and a maximum number of direct links for customers. This
service pattern is condensed into the Service Intention (SI, see Figure 1), which is the functional
requirement for the timetable development.
With the help of origin-destination matrices, an offer of train services with lines and frequencies is
developed to meet the customer needs. Since at this point it is not known whether this offer is feasible it
is called Service Intention rather than Service Offer.
The Rail 2000 service concept consists of
Train-lines and frequencies specifying the customer-relevant information, such as stop stations,
interconnection possibilities for each service type,
Arrival and departure platforms at stations and
Rolling stock.
The basic principle with respect to journey times is as fast as necessary to enable good connections in
the next hub. All-day regular departing and arrival times for each service type facilitate the use of public
transport, in particular for occasional users.
In the second step, the feasibility of this service intention is checked by generating a feasible schedule. If
this is possible, a schedule is provided as proof of feasibility, otherwise both steps have to be repeated
until a feasible service intention is found.
The design of this timetable takes into account mixed traffic. It attributes on the same network systematic
slots for international, national and local passenger trains as well as national and international (transit)
freight trains.
The desired regular interval timetable is characterised by
parallel time-distance graphs
symmetry of arrival and departure times around the defined symmetry time of the system
Effects of such a systematic timetable are
the creation of systematic transfer connection groups in hubs
the crossing of two given trains occurs always at the same place and the same minute of a full hour

an increase in market demand due to the regularity of the service

higher productivity through a systematic production process
Optimisation of rolling stock usage through systematically minimised terminus times

Once the timetable is finalised, it is converted into the production plan which consists of the timetable
and in addition all the detailed resource assignments needed for the daily production. If due to some
incident a certain required resource is missing, the production plan has to be adapted in a way that the
intended timetable still can be realised. If this is not possible, the service intention provides the objective
for designing an alternative timetable in real time which comes closest to the original promise made to the
customer.
2
The Service Intention
Consider the example of connecting trains and minimizing delays to customers as a measurement of
service level: does providing a high level of service entail ensuring that connecting trains always wait for
late trains? Or does it entail ensuring that connecting trains never wait for late trains? Or perhaps it
involves a combination of both strategies depending on the individual circumstances?
2.1 Problem formulation: an Example
Consider the following scenario:
Train A from Zurich to Lucerne carrying 500 passengers is running 10 minutes late. All the
passengers on this train require a connection to Arth-Goldau provided by train B at Zug.
Train B from Lucerne to Arth-Goldau currently has 200 passengers on it and they are waiting to
depart in 2 minutes from Zug.
Train C provides the same connection to Arth-Goldau from Zug, but it departs 15 minutes later
than train B.
A possible strategy for the aforementioned scenario might be as follows: ensure that the lowest possible
number of passengers have to endure a delay. In this case, train B will wait for train A, since only 200
customers will be delayed instead of 500.
However, consider the situation where the commercial value of train B leaving on time is far greater than
the commercial value lost by further delaying the 500 passengers from train A, e.g. if train B provides
further critical connections at Arth-Goldau that have no substitute. It is still possible for train A passengers
to use train C to reach their final destination within a 5 minute delay threshold, but customers waiting for
train B at Arth-Goldau could be subject to a greater delay. These are, of course, just 2 possible scenarios
out of a multitude of possibilities especially when you consider the complexity of even a small segment of
the entire railway network.
The given example shows that taking operational and planning decisions aimed at providing customers
with a high level of service based on varying criteria is not a straightforward process there are several
variables involved in making an optimal decision such as passenger satisfaction, commercial
considerations, and overall stability of the schedule.
Firstly, the importance of particular train services must be identified; secondly the impact on the overall
customer service (and inherently the loss of associated value) when particular train services are
cancelled must be considered; thirdly, it should be possible to evaluate whether proposed services (and
subsequent changes to these services) are realisable with the given resource constraints (rolling stock
and topology).
The notion of the Global Service Intention (GSI) is built upon several other concepts that need to be
defined first. These are presented in the following section.
2.2

Basic Definitions for functional demonstrator

1. Train Service Intention: describes a train service from a start station to an end station and
consists of the following:

a unique identifier;
an importance value (realistically it should be a combination of various factors such as
passenger loads and commercial importance for example);
a list of stations served by the service intention;

a minimum transition (duration) time for the service intention from start station to end station;
a list of connections that the service intention provides and at what stations the connections
take place i.e. a list of other service intentions that this service intention provides a
connection for and the stations at which these connections take place.
the timeframe (period within the hour) in which the service intention originates at the start
station and the timeframe in which the service intention has stops at stations along the way.

2. Local service intention (LSI): the set of service intentions corresponding to a particular station
region. For example, the set of service intentions available from the station of Lucerne.
3. Global service intention (GSI): the collection of all service intentions on the railway network.
4. Stability: a timetable is considered stable when it is possible to absorb a specified set of degraded
resources while continuing to fulfill the GSI.
5. Iterative Control Algorithm (ICA): a scheduler that can derive a concrete timetable based on the
GSI. See Figure 3 and [1] for further information.
2.3 Graphical Representation
The data structure (model) used for the GSI is shown in the following example (see Figure 2)
Luzern Ebikon Buchrain Lngenbold Gisikon-Root Rotkreuz Chmleten
0:0

Zythus

Cham Alpenblick

Chollermli

Schutzengel Zug

Lindenpark

Neufeld

Baar

3.75
3.75

0:7.5

AA

0:15

0:22.5
0:30
B

0:37.5

BB

0:45

eB
B

C
C

IR_D

IR_D
IR_D
E

c1
C1

1:7.5

1:15

0:52.5
1:0

IR_D
E

1:22.5

FF

1:30
1:37.5
1:45
1:52.5
2:0

Figure 2: Example of GSI graphical representation

The data structure is a space-time graph with the following properties:

The time dimension is represented as a sequence of temporal layers in Figure 2. For the purposes of
a functional study, an hour is broken up into layers of 7.5 minutes each (this is a parameter).
Experience has shown that 7.5 minutes is the typical threshold at which a customer will tolerate
waiting for a train, so it makes a good choice when designing a GSI as deficiencies in service
intentions can quickly be exposed.

Each station is represented as a set of nodes. One node per station exists in each of the temporal
layers. Within each node is defined a list of connections of service intentions and a train change time

for customers. A connection example is shown as a blue line labeled c1 in Figure 2 connecting
service intentions C and IR_D at station Rotkreuz.

Stations are classified as either Main stations or Small stations:

o

Main station: a station that lies exactly on the centre of the temporal layer (e.g. at 0 mins, 7.5
mins, 15 mins, etc.). These stations are considered to be the main system nodes of the
railway network. This means that if a service intention originates or stops at one of these
stations, it has a restriction where the scheduler (ICA) must ensure that in the final timetable
the train must be at the station within a time equal to or within a certain threshold (which is
3.5 minutes in the given example of Figure 2) of the relevant centre of the temporal layer. For
example, in Figure 2 there is a service intention named IR_D going from Baar to Lucerne.
The service intention is at centre 0 mins of the 9th temporal layer at node Zug and ends at
node Lucerne at centre 22.5 mins of the 12th temporal layer. This means that the scheduler
is restricted to producing a timetable entry for this service intention that is 0.0 mins +- and
22.5 mins +- at those particular stations. Of course, the transition time for the service
intention must also be valid and enforced for this process to work. An example of a final
timetable entry derived from this service intention could therefore be: Zug (01:02) Lucerne
(01:25) with a transition time of 23 minutes.

Small station: a station that lies on the boundaries of the temporal layers. Unlike the main
stations, during timetable generation there is no restriction on the times that the scheduler
(ICA) must assign to trains starting or stopping at these stations in the timetable. In case of
operative rescheduling, only delays are allowed as departure times have been committed in
the published timetable.

Service intentions are represented as directed edges connecting nodes within or across temporal
layers in the graph (depending on the transition time length). The edges store all the pertinent
information about the service intention (a unique identifier, importance value, transition time) and in
conjunction with the implied information obtained from the node structure of the GSI, other
information that is defined is: the timeframe of the service intention (i.e. the temporal layers through
which the service intention edge passes through) and the list of stations served.

The GSI is therefore a high level representation of a final timetable. This abstract representation can be
used to derive a final timetable as described in section 1.2. It can also be used to provide useful
information on the effects of disruptions and cancellations of train services on the customer.
In [7], V. Mahadevan proposes to use the transitive hull as a superior method of relaxing the GSI versus
using simple value calculations, recalling that service intentions can possibly provide connections to other
service intentions. So cutting a service intention from the GSI can potentially affect other service
intentions as well. Therefore simply subtracting the value of a standalone service intention is not the most
accurate measure of determining the value lost when the GSI is relaxed. The collection of service
intentions and their associated values that are affected by a service intention cut should therefore be
considered when relaxing the GSI. The transitive hull of the GSI captures these relationships; when a
service intention is removed from the GSI, the transitive hull can be recomputed and compared with the
previous transitive hull to ascertain the value lost across the collection of service intentions. However, in
this case an importance value must be assigned to each connection as well so that it can be determined
what value is lost explicitly due to the connection loss in addition to the value loss caused by removing
the standalone service intention.
It is also possible to simply specify a static set of rules for GSI compromises based on predefined criteria.
For example, one could specify that a certain set of service intentions must never be removed from the
GSI in a static rule set or have a linear ordering of service intentions to be relaxed.
3
Timetable Generation Process
Once the GSI is defined in a way similar to the example of 2.1, it is straightforward to decompose it into a
set of Local Service Intentions (LSI as defined in 2.2). These LSIs together with the available resources
and the geographical constraints serve as input to either the planning process where the planner
endeavours a conflict-free solution for the timetable (TT) or the dispatcher tries to resolve a conflict in real
time and to reschedule the original TT.

In both cases the resulting TT has to consider the available resources and to satisfy all the given
constraints for the local TT. In the Timetable-Planning process where the planner generates the TT for a
whole timetable period, it is assumed that at production time the required resources will be available out
of a certain stock containing predefined quantities of each production resource such as rolling stock,
personnel stock, track capacity, etc.
In the last step Production Plan of the design process chain of Figure 1 the TT is transformed into a
Production Plan where concrete resources out of the stock are assigned to the individual trains on a daily
basis. In case of the real-time rescheduling of the TT it is usually a particular resource-assignment which
is no longer valid due to a certain incident like for instance an occasional track speed restriction or a
locomotive with a machine malfunction.
In this case the actual TT has to be adapted in such a way that the rescheduled TT is still conflict-free with the new
resource assignments and either the given GSI is still valid or only minimally compromised (back-loop in Figure 3
if SI is not satisfied).
Iterative Control Algorithm (ICA)
Global SI (incl.
Quality Criteria)

Local SI
Local
Local SISI

Available
Resources
(rolling stock,
topology)

Find global
tentative TT

Check feasibility
locally on detailed
topology

Detect conflict,
Add constraint
NO

Time/Distance
Constraints

YES

TT constrained?
additional
Constraint Report

Compromise the GSI

(evaluate punishments for
failing to satisfy offers)

TT Feasible?

NO

Conflict-free
TT

SI satisfied ?

Timetable
YES

Figure 3: Process overview of timetable generation using the ICA

3.1 The PULS 90 network decomposition approach

In order to divide the planning problem into smaller, more manageable and resolvable constituent units,
the infrastructure network is segregated during the development of the GSI into major junction areas (big
stations with their inlet areas which are called condensation zones). In these locations there is almost no
spare capacity (due to functional reasons such as the high number of line connections, and other
resource allocations etc.). The surrounding parts of the network, with train travel time reserves, are called
the compensation zones (see also [9]).
Within the condensation zones the speed of the planned track slots is almost identical to the maximum
permitted speed. Hence, within these areas, minor differences between the travel speeds of different
trains can then be ignored when determining track slot utilisation. This effect is captured in the concept of
the PULS-Pattern which is specifically designed for each condensation zone (see Figure 4).
Because of the comparatively small amount of possible operationally viable combinations in the PULSPattern (see Figure 4 below, left side) compared to a continuous search space as in a normal timedistance diagram (see Figure 4, right side), it becomes possible to make a quick manual assignment of
trains to track slots and to suitable infrastructure elements. Additionally, this also makes it possible to
rapidly calculate new schedule scenarios. Conflicts and predetermined dependencies between the trains
can easily be detected and managed. The PULS-concept is described in detail in [8]).

Figure 4: PULS-Pattern Diagram with track slots (left side) and a normal Time Distance Diagram showing the
same track slots (right side)

On the long stretches of track in-between these major junction areas the planed track slots are supposed
to have enough reserves so that the train can still reach the intended arrival time of its required slot in the
neighbouring junction area, even if small perturbations occur.
While the major junction areas have no temporal reserves, and therefore can only be controlled in terms
of route (re-)assignments, trains in the compensation zones can be accelerated or slowed down
according to their actual slot assignment in the adjoining saturation zone that they are heading to or from.
3.2

Automatic Generation of train schedules

A computational test environment for the automatic generation of periodic timetables for a given GSI
utilising PULS-patterns in condensation zones has been developed and tested at the ETH Zurich.
Figure 3 shows the process diagram for the algorithmic implementation (ICA).
The goal is to create detailed train schedules, which specify an exact itinerary through the railway
topology with passing times for each train. This way the provided timetable is guaranteed to be conflictfree, i.e., assuming no delays, all trains can run exactly as planned without creating safety conflicts.
As it appears intractable to consider the detailed topology all at once, a two-level approach for generating
conflict-free train schedules is used. In the macroscopic (or macro) level, given a GSI, an abstraction from
the detailed track topology for creating a draft timetable is used. In the microscopic (or micro) level,
starting with the draft timetable from the macro level, detailed train schedules are constructed by
considering locally precise topologies, the corresponding safety system as well as accurate train
dynamics.
The macroscopic level can be modelled as a Periodic Event Scheduling Problem (PESP, see [5]) whose
output (departure and arrival times) serves as the input for the micro level to check feasibility by finding a
feasible routing. Constraints of the PESP model are the travel times of trains between portals, dwell times
in a station, connection constrains, headway constraints as well as other minor constraints. The classical
PESP model with fixed passing times, however, is quite restrictive and could lead to infeasibility at the
microscopic level, especially in condensation zones where the speed profile is fixed. To allow for more
flexibility, an augmented PESP (Flexible PESP, or FPESP) was developed that assigns time slots (lower
and upper bounds) instead of exact passing times. This additional flexibility can be then exploited when
scheduling in the micro level to increase chance to find a feasible solution. Results show that it is possible
to generate macro schedules for scenarios like the whole central Switzerland in few minutes, where each
train has an average flexibility of 8 minutes [3].

In the microscopic level the new scheduling approach that decomposes the problem geographically into
condensation and compensation zones [1] was applied. Different policies for generating micro train
schedules can then be applied to the two zones according to their distinct properties.
As condensation zones are expected to have high traffic density, they comprise bottleneck resources that
should be occupied as shortly as possible. Therefore, trains are required to travel through the
condensation zones with maximum speed, i.e. no time reserves are included. Thus, the speed profile is
completely fixed and it is sufficient to assign one passing time per train (e.g. at the portal or at the
platform) from which all other passing times within the condensation zone are derived, once the itinerary
has been fixed. This passing time is chosen from the provided time slots of the FPESP that lead to the
optimal local schedule fulfilling the given local service intention.
On the other hand, in compensation zones there is a large variety of speed profiles for the train to
connect the two main stations. These zones will be scheduled after the condensation zones. Their portal
passing times, which are now fixed by the scheduling in condensation zones, can be matched into a
conflict-free schedule by optimising the quality of the speed profiles as the main degree of freedom in the
conflict-free solution.
The crucial aspect of microscopic scheduling is the guarantee that the train schedule is conflict-free. Two
trains are in conflict if they allocate the same topology resource element within a certain safety time span.
This is typically modelled by a conflict graph [10]. A conflict-free schedule therefore corresponds to an
independent set of maximum cardinality in the conflict graph. Hence, an alternative model, called
Resource Tree Conflict Graph (RTCG, [2]) was required, which is based on an integer multi commodity
flow formulation that incorporates the railway topology as well as passing times and speed profiles of the
trains. Train itineraries are represented in a tree structure to identify the location of conflicts. For each
track element the train allocation times are gathered in sets that contain all train itineraries competing for
the resource at the same time. Whereas the structure of the conflict graph formulation is usually weak,
substantially stronger constraints can be generated by considering all trains simultaneously (instead of
just two) for each resource. As a consequence, the resulting integer program is then solved with a
commercial solver much faster, as illustrated in Table 1.
Scenario
Lucerne, service intention
2007
Berne, condensed
hypothetical LSI

# trains
in 30
24

# conflicts
conflict graph
2600

Solution time
conflict graph [s]
1

# conflicts
RTCG
1500

Solution time
RTCG [s]
0.1

27

7 400 000

2490

5200

Table 1: Run Time Performance in Local Timetable calculation for Lucerne and Berne, which has a much more
complex topology and hence a bigger solution space

More detailed information concerning the ICA can be found in [1], [2]and [3].
4
Test environment for Local Service Intention in Lucerne
In order to proof the practical application of the SI-concept outlined above, a pilot project has been set up
in the area of Lucerne.
Whereas the GSI and the global timetable describes the service offered on the global SBB-network and is
today under responsibility of 5 different operations control centres (Betriebsleitzentrale), each controlling
a certain region (see Figure 5), the LSI of the area of Lucerne (see dotted line in lower left of Figure 5) is
controlled by the traffic control centre Lucerne (Fernsteuerzentrum, FSZ) from where all relevant
interlockings (e.g. the red circle of the station of Lucerne) in the area are remote-controlled.

Timetable /GSI
Disposition/GSI

Betriebsleitzentral
Betriebsleitzentrale
Mitte
e Betriebsleitzentrale
Region
Betriebsleitzentrale
Mitte
Region
Betriebsleitzentrale
Mitte
Region
Region Mitte
Region Mitte

Operations/LSI

FSZ/Lucerne
FSZ/Luzern
FSZ/Lucerne

Safety
Interlocking/Bahnhof
Interlocking/Bahnhof
L
Interlocking/Station
L
Figure 5: Hierarchical organisation of network train operations in SBB

Thus, in a first pilot phase the intention is to control the objectives of the operations in a local network
area which in the organisational structure corresponds to the level of the FSZ Lucerne.
Figure 6 illustrates the conventional information and control flow with the control level of the FSZ Lucerne
highlighted.
Timetable
Development

Disposition

Operation

TOC

Yearly

Yearly

Operation
Control

Train
state &
position

Train

Telephone

Timetable

Safety
System

C
U
S
T
O
M
E
R

Online

Yearly, daily adjusted

Traffic
Control

High update frequency

Medium update frequency
Low update frequency

If correct
information
is available
in time

Customer
Information
System

Figure 6: Conventional Information and control flow

One of the major disadvantages of the existing situation is that there is only little communication between
the Disposition and Operation level on the one hand and the Operation and the Train Operating Company

(TOC) on the other hand. Furthermore, there is no clear dispatching objective that the disposition and the
operation level can agree on when operational disturbances occur. As a consequence, dispatching
decisions are often ad hoc and in case of larger problems, there is no clear guideline for the problem
resolution. In such situations it is hard to find a temporary solution that can be clearly communicated to
the relevant actors and the customers involved.
Compensation zone

Condensation
zone

Traffic Control
Room Lucerne
Gtsch
FST-LT90
TCIS
Seetal

Lz
Station
TCIS
Entlebuch

Lz Station
(backup)
Operation
Assistant

Information
Assistant
Shift
supervisor

Figure 7: Test perimeter and associated operations controllers in the traffic control centre (TCIS: Train control
and Information System)

Figure 7 illustrates the geographical test area together with the line responsibilities in the traffic control
room. During the pilot phase, the new role of the Rescheduling Officer (RO) will be assigned to the shift
supervisor who takes the responsibility for the LSI. In order to be able to make rescheduling decisions
based on the LSI he will be equipped with the Rescheduling Framework (RFW) which is designed to
provide him with all required information.
Each time a schedule deviation occurs, he receives a notification together with the information about the
type and the magnitude of the deviation from the actual planned system state. Based on this information
he tries to find a new conflict-free plan basically using a PULS-user interface (presenting a display, similar
to the left part of Figure 4) that allows him to modify the track slot properties according to the actual
resource availability and to change the spatio-temporal assignments of the different track slots until all
existing conflicts including track occupation and required connections etc. are resolved. Figure 8 shows
the information flow of this rescheduling process.

10

Initial
Timetable
extract

Reverse
Engineering

LSI (incl. Quality

Criteria)

Rescheduling Officer FSZ Lucerne

Manual attempt to find conflict-free
local TT using new Rescheduling
Framework

Local SI

(Constrained) available
Resources
(rolling stock, topology)

LSI satisfied ?

Time/Distance
Constraints

YES

NO

Compromise the LSI

Rescheduled Timetable

Figure 8: Process overview of the adapted iterative timetable generation initialised with a reverse engineered
LSI and using the Rescheduling Framework to reschedule the timetable and/or relax the LSI constraints
respectively

As soon as the new schedule is committed by the RO, the corresponding changes to the production plan
are generated automatically. The production plan contains the new instructions and information for all
other actors involved, including train drivers, conductors and the different operators in the traffic control
room. By programming the interlocking (using the TCIS) the operators activate the required adaptations
to the signalling. The information assistant informs the customers by appropriate announcements either in
the station or in the trains through messages to the conductors. The train drivers are instructed to alter
their speed if necessary in order to avoid stops in front of red signals.
The enhanced system environment for the pilot phase in Lucerne is indicated in the following diagram:
Disposition

Timetable
Development

TOC

Operation

Yearly

Yearly

Operation
Control

Timetable

Train
state &
position

Safety
System

C
U
S
T
O
M
E
R

Online

Traffic
Control

Telephone

Train

Rescheduling

LSI
Yearly, daily adjusted

Tolerance
thresholds

Customer
Information
Systems

Rescheduling
Frame work

Real time
scheduler
Train Driver &
Customer
information

High update frequency

Medium update frequency
Low update frequency

Online

Online

Figure 9: Information and control flow in test setup of the field test in Lucerne (see also [6] for further details)

Several improvements can be noticed. Firstly, the LSI serves as an objective for dispatching decisions
which is commonly agreed between the disposition and the operation level. On the other hand with the
help of a threshold detector, the RO is much better informed about intolerable deviations from the current
production plan. Trigger conditions for executing a rescheduling can be made explicit. Due to the better
decision support provided by the RFW, the RO can react faster and more efficient than before.

11

Already during the first low level implementation of the pilot environment it shows that, in comparison the
conventional situation, the TOC and the customer are better integrated in the information and control flow.
This is achieved mainly through the online communication to the train driver, the conductor and the
customer. What may be the most important improvement of the RFW is that it supports operation to
ensure that most of the time there is a valid schedule available which serves as a basis for providing all
actors with the required information, and in time. When technical problems arise, being able to provide
the customer with the right information about an alternative connection to reach his final destination in
time is an important benefit.
Another decisive factor for the required high precision in controlling the traffic flow is the efficient
communication to the train driver. Therefore the system environment of the pilot contains an online
connection to a dedicated cabin display. Information about new track speeds is communicated to the train
driver by transmitting the deviation of the actual time at the current train position from the time which was
planned according to the actual schedule. Technically this is achieved by measuring the current train
position on board and comparing it with the planned position at the actual time. The latter is first
transmitted to the on board unit before the train trip starts and is updated each time that there is a change
in the time or route information of the appropriate track slot due to a rescheduled production plan.
In addition to the time deviation in seconds, the on-board unit contains information about the
recommended driving mode (e.g. a accelerate, decelerate, keep speed etc.) and a recommendation
about where within the 8 minutes trip time ahead the existing time deviation should or respectively can be
compensated. The possible time compensation (considering technical and safety restrictions) is indicated
as white bars whereas the recommended time compensation is indicated as grey bars. Positive values
indicate the amount of seconds which have to be made up for (i.e. when the train is late) and negative
values indicate that train has to lose time (i.e. when he is ahead of schedule). Figure 10 shows an
example for a situation of a slightly delayed trip.
Legend
[t]
Next 8 minutes trip time ahead
[t]
running time deviation in seconds
Total time deviation with regard to
schedule
Maximal recoverable delay in
corresponding time window
Recommended correction in
corresponding time window
Indication field for train number,
origin, destination and intermediate
station
Driving mode: example for
intermediate fast running
Figure 10: Driver display. Example for a slight delay

5
Outlook: From Local to Global Service Intention
Within the first three chapters of this article we tried to outline the concept of the GSI as the functional
objective for controlling the network traffic in SBB. In chapter 4 we described the first pilot implementation
of an LSI-based operating environment. With the installation and practical test of the RFW in the traffic
control centre in Lucerne, we intend to provide a proof of concept of the LSI-based dispatching approach.
Our test cases are small timetable constellations in which comparatively small operational problems are
encountered and that can be solved manually with the help of the RFW. However, these cases help us to
learn more about the single elements of the concept so that we can increase the problem scope stepwise
in the future. As soon as our experience with the SI and the RFW has been increased, we intend to
extend the test scenarios in such a way that also the disposition level and correspondingly larger portions
of the network can be integrated into the test problems. Therefore we need to trial several condensation
zones (e.g. Olten and Arth-Goldau in addition to Lucerne) in order to gain experiences with the GSI. The

12

theoretical work described in chapters 2 and 3 will be the basis for future developments and extensions of
the RFW so that larger problem scenarios, which in the pilot phase lead to a fall-back to conventional
methods, can also be handled by the RO. The clear distinction between the Disposition and Operation
level must be resolved in the future. Whilst the global part of the concept will be subject to deeper
investigations, the local part can be put into production in the near future.
6
Conclusions
Never before, the objective of real time rescheduling during operation has been focused on reducing the
overall customer annoyance, taking total trip duration into account. To achieve this, it is necessary to
keep the production plan in sync with the actual operation even if irregularities and disturbances occur.
This is achieved by monitoring travel and departure times of trains and resource states in general and
rescheduling trains in case of exceeded tolerances.
On the other hand train drivers are empowered to keep to the required tolerances by receiving all the
necessary information such as travel time and the admitted tolerances constantly on a special display.
This allows them to compensate for deviations.
The ongoing implementation of the Bahn 2000 process chain results in a consequent integration of
planning and production methods which allow keeping operation of railway traffic under control of service
intentions.
Acknowledgements
We would like to thank our colleague V.Mahadevan for providing us the example data and a prototype
implementation of the GSI demonstrator, as well as A.Smith for his support with the English language.
References
[1] Caimi G., Burkolter D., Herrmann Th., Chudak F., Laumanns M. Design of a railway scheduling
model for dense services, Journal of Networks and Spatial Economy, 2008
[2] Caimi, G., Chudak, F., Fuchsberger, M., Laumanns, M., Solving the train scheduling problem in a
main station area via a resource constrained space-time integer multi-commodity flow, Technical
report, Institute for Operations Research, ETH Zurich, 2007.
[3] Caimi, G., Fuchsberger, M., Laumanns, M. and Schpbach, K., Periodic Railway Timetabling with
Event Flexibility, In C. Liebchen et al, editors, ATMOS 2007 7th Workshop on Algorithmic Methods
and Models for Optimisation of Railways, Schloss Dagstuhl, 2007
[4] Laube F., Mahadevan V., Bringing costumer focus into every nut and bolt of the railway: Swiss
federal Railway's Path into the future, WCRR Seoul, 2008
[5] Liebchen, C., Mhring, R., The Modeling Power of the Periodic Event Scheduling Problem: Railway
Timetables - and Beyond, In F. Geraets et al, editors, ATMOS 2004 4th Workshop on Algorithmic
Methods and Models for Optimisation of Railways, Springer, 2004.
[6] Luethi M., Nash A., Weidmann U., Wst R. Increasing Railway Capacity and Reliability through
Integrated Real-Time Rescheduling, WCTRS Berkeley, 2007
[7] Mahadevan V. (2007), Describing and Evaluating Train Services on the Swiss Railway Network from
a New Perspective, Conference paper STRC, Monte Verit / Ascona, 2007
[8] Roos S, Bewertung von Knotenmanagement-Methoden fr Eisenbahnen, Masters thesis at the
Institute for Transport Planning and Systems, ETH Zurich, 2006
[9] Wst R., Dynamic rescheduling based on predefined track slots, WCRR Montreal, 2006.
[10] Zwaneveld, P.J., Kroon, L.G., Romeijn, H.E., Salomon, M., Dauzre-Prs, S., Van Hoesel, S.P.M.,
Ambergen, H.W., Routing Trains through Railway Stations: Model Formulation and Algorithms,
Transportation Science 30(3), 1996.

13