QUT Digital Repository:

http://eprints.qut.edu.au/

This is the author version published as:

This is the accepted version of this article. To be published as :

This is the authors version published as:

Goodman,C.J.andSiu,L.K.andHo,T.K.(1998)Areviewof
simulationmodelsforrailwaysystems.In:International
ConferenceonDevelopmentsinMassTransitSystems1998,2023
April1998,London.
Catalogue from Homo Faber 2007

Copyright1998IEEE

A REVIEW OF SIMULATION MODELS FOR RAILWAY SYSTEMS

C.J. Goodman

L.K. Siu

T.K. Ho

University of Birmingham, UK

Kowloon-Canton Railway Corporation, HK

Hong Kong Polytechnic University, HK

1.

INTRODUCTION

With the advances in computer hardware and software

development techniques in the past 25 years, digital
computer simulation of train movement and traction
systems has been widely adopted as a standard
computer-aided engineering tool [1] during the design
and development stages of existing and new railway
systems. Simulators of different approaches and scales
are used extensively to investigate various kinds of
system studies. Simulation is now proven to be the
cheapest means to carry out performance predication
and system behaviour characterisation.
When computers were first used to study railway
systems, they were mainly employed to perform
repetitive but time-consuming computational tasks, such
as matrix manipulations for power network solution and
exhaustive searches for optimal braking trajectories.
With only simple high-level programming languages
available at the time, full advantage of the computing
hardware could not be taken.
Hence, structured
simulations of the whole railway system were not very
common. Most applications focused on isolated parts of
the railway system. It is more appropriate to regard
those applications as primarily mechanised calculations
rather than simulations.
However, a railway system consists of a number of
subsystems, such as train movement, power supply and
traction drives, which inevitably contains many
complexities and diversities. These subsystems interact
frequently with each other while the trains are moving;
and they have their special features in different railway
systems.
To further complicate the simulation
requirements, constraints like track geometry, speed
restrictions and friction have to be considered, not to
mention possible non-linearities and uncertainties in the
system.
In order to provide a comprehensive and accurate
account of system behaviour through simulation, a large
amount of data has to be organised systematically to
ensure easy access and efficient representation; the
interactions and relationships among the subsystems
should be defined explicitly. These requirements call
for sophisticated and effective simulation models for
each component of the system.
The software
development techniques available nowadays allow the
evolution of such simulation models. Not only can the
applicability of the simulators be largely enhanced by

advanced software design, maintainability and

modularity for easy understanding and further
development, and portability for various hardware
platforms are also encouraged.
The objective of this paper is to review the development
of a number of approaches to simulation models.
Attention is, in particular, given to models for train
movement, power supply systems and traction drives.
These models have been successfully used to enable
various what-if issues to be resolved effectively in a
wide range of applications, such as speed profiles,
energy consumption, run times etc.

2.

TRAIN MOVEMENT

Train movement is the calculation of the speed and

distance profiles when a train is travelling from one
point to another according to the limitations imposed by
the signalling system and traction equipment
characteristics. As the train has to follow the track, the
movement is also under the constraints of track
geometry and speed restrictions and the calculation
becomes position-dependent. The method of arranging
the data representing the track geometry and speed
restrictions is therefore critical to achieve effective and
swift simulation.
2.1

Track-based data representation

In early simulation software designs [2], the data was

stored in two-dimensional array with the rows of the
array denoting the signalling blocks with fixed block
signalling. The track-based data include block identity,
gradients, speed restrictions, coasting points and signal
aspects etc. within the signalling block. The major
advantage of this structure is easy referencing.
However, only a single data type (e.g. floating-point
numbers) is allowed within the array, which is inflexible
for accommodating the diversified nature of data.
Besides, the array structure does not offer any
representation of track layout except that adjacent rows
in the array may depict a sequence of adjacent signalling
blocks. When it is necessary to describe the track
connections within a complicated railway network, the
array structure must be enhanced or additional data
structures are needed. Further, the size of the array
required varies with application, a supposedly large

enough array is usually defined. Thus, excessive

memory space is reserved in the simulator in order to
meet the demands of most applications.
An object-oriented software approach [3] provides the
solution for the above deficiencies. The object-oriented
concept allows a set of objects in the physical world
sharing similar properties and performing similar
operations to be grouped together in one class. A class
specifies a data structure for its objects and a number of
permissible operations. An objects data can only be
accessed and modified by the permissible operations.
This protection of data from arbitrary and unintended
use and access of data is called encapsulation.
The object-oriented concepts have been applied in
railway network modelling [4] [5]. The structure of the
network is represented by a number of node objects
joined by link objects. The nodes can be used to
represent stations, junctions, points and termini etc.
whilst the links are the tracks connecting the above
features.
They have their own data structures
characterising the corresponding objects, and the data
structures accept mixed data types for various types of
data. Train movement is realised by moving the train
from one node to the next through a permitted link,
which contains the necessary information for the
movement calculation. Creation and initialisation of any
new objects from a class can be done at run-time and no
reserved memory is needed.
The properties of inheritance and polymorphism for
object-oriented systems are the key factors to achieve
efficient software coding.
Inheritance is an
implementation of generalisation as a new class can be
defined by the definition of an existing class. It makes
the data structure and operations of an existing class
(superclass) physically available for re-use by the new
class (subclass); hence it enables code sharing.
Polymorphism allows modifications of operations in a
subclass so that different operations on different objects
are possible in a hierarchy classes.

Despite its simplicity, a time-based model usually makes

a high computational demand as a significant amount of
information has to be produced within each time update.
Even though computation effort can be alleviated by the
choice of larger time update intervals, the attainable
level of details is then compromised. This high
computational demand can only be justified in
applications where full details of every move of the
trains are needed, such as energy consumption and
signalling design studies [6] [7].
An event-based model, on the other hand, denotes the
progress of train movement by the occurrence of a
sequence of pre-defined events, such as arrival at and
departure from stations [8] [9] [10]. Since the events
are linked to each other according to the interactions
among trains through the signalling, power system and
other system characteristics, one event, as the
consequence of a previous event, will trigger or cause
another event to happen. As a result, a chain of events
determines the progress of the trains.
Computational effort can be substantially reduced
because the calculation of exact details of train
movement between pairs of events is skipped.
Nevertheless, this apparent advantage may be
overshadowed by the fact that the passage of time is
irregular and the updates of train movement are not
carried out synchronously. It is possible that the
processing of an event has to be postponed because the
event to trigger it has not occurred in time, or it is
processed first on the assumption of certain conditions
and it will be re-processed if the assumption is found
invalid later. Therefore, great care is needed in the
development of event-based models in order to avoid
the above drawback.
Event-based models find
applications mainly in traffic control and timetabling
studies [11] [12], in which only the information of
timings at certain events are of main concern and quick
simulation results are expected.

3.
2.2

Movement calculation

Depending upon the level of detail required in the

simulation studies, there are two major approaches to
calculate train movement as a function of time, timebased and event-based models.
In time-based simulation models, the time span is
divided into evenly-spaced intervals and the train
movement is evaluated at each interval. Time is
considered to be frozen whilst the behaviour of the
trains is updated simultaneously and the system evolves
continuously with time. This is conceptually close to
how the trains move along the track in reality, hence it is
easier to design and build simulation models with the
time-based approach.

TRACTION POWER SUPPLY AND DRIVE

SYSTEMS

In designing new traction power systems, there is a

considerable range of alternatives to be contemplated.
Traction power engineers would need to analyse and
compare the performance of alternative proposals to
arrive at a cost-effective design which satisfies the
clients requirements. This calls for software procedures
for solving the power network equations repeatedly in
order to establish a complete picture of power demand,
energy consumption, voltage and current of the feeder
stations and trains. Due to the iterative nature required
for solving the network equations, efficient algorithms
are often needed for the power network simulator to
provide accurate results within a reasonable computation
time. There are three areas to be considered in
developing a traction power network simulator:

the representation of traction power networks in the

solution domain;
the train position locator; and
the power network equation solver.

The representation of traction power networks in the

solution domain has a direct influence upon the way in
which the power network problem is formulated, and
hence the train position locator and power network
solver organised. There are, by and large, two types of
approach to solve the traction power network problem
namely:

3.1

modified load flow type approach [14] [15] [16];

and
direct matrix method combined with piece-wise
linearised circuits approach [1] [2].
Modified load flow approach

For the modified load flow type calculations, the

traction power network simulator is quite often
separated from the train movement simulator (or train
performance calculator). The traction power network
simulator, as a stand-alone module, takes in the train
movement results, such as train locations and train
power demands etc. from data files or intermediate
stores. This approach, from a programming viewpoint,
provides a much easier interface between the train
movement and traction power simulators. It does,
however, provide no direct reflections of voltage
variations back to the train movement calculations. With
modern three-phase drives, the traction drive is less
dependent on supply voltage than is the case with DC
motors, but all drives do have a designed graceful
degradation response to reduced traction voltage. Thus,
if the performance limits of the total system are to be
properly examined, feedback of the power network
solution to the traction performance calculations
becomes essential.
Another feature of the modified load flow type approach
is the combined network solver for AC/DC networks,
e.g. 750 V DC to 11/33 kV AC ring. The equations of
the two networks are alternatively solved until the final
solution of the network is obtained for a given time
update period. This particular feature provides the
engineers a useful tool to analyse and optimise the
complete AC/DC system. In the network equation
solver, standard load flow procedures using the
conventional algorithms, such as Gauss-Seidel and
Newton-Raphson methods, need to be modified to cater
for the non-linearities of the traction systems, such as regenerative trains, system non-receptivity, etc. These
however apply to the direct matrix approach as well.
3.2

Direct matrix method

For the direct matrix approach, the network matrix can

be formulated using either mesh analysis or nodal
analysis. Comparing the two approaches, the nodal
analysis is more suitable for dealing with complex
networks, because it is often easier to identify nodes
than loops in non-planer networks. From a programming
perspective, automatic network set-up procedures and
advanced network graph techniques are easier to
implement with the use of the nodal approach. Fig. 1
shows the circuit representation of a typical DC traction
power network with a branch. It is not difficult to see
that the railway traction power network is characterised
by sections of ladder type networks infrequently crossconnected. For this characteristic topology, the sparse
matrix technique coupled with efficient matrix
elimination methods leads to an expeditious network
solution, since it does not suffer from the fill-ins that the
direct inversion of the coefficient matrix requires.
3.3

Static and dynamic ordering

General speaking, the matrix of the traction power

network equations is of positive definite (PD),
symmetric and sparse type.
Many sparse matrix
elimination techniques tailored for PD matrices may be
used, e.g. LLT decomposition, Cholesky decomposition
etc. For the sparse techniques, the essence is the
equations ordering. There are generally two types of
ordering method namely, static and dynamic ordering.
Typical efficient examples for long and thin ladder type
networks include the Cuthill and McKee algorithm [17]
and reverse Cuthill and McKee algorithm [18] which
make use of the property that the zero elements situated
before the first non-zero element on any row always
remain zero, and take the advantage of the variation of
the matrix bandwidth, also referred to as the matrix
envelope.
For dynamic ordering, the minimum degree algorithm
provides an efficient alternative for solving the network
matrix. The minimum degree ordering scheme is one in
which the pivot selection is made in accordance with the
way in which the coefficient matrix develops, rather than
simply from the structural properties of the original
matrix. This is actually a heuristic algorithm for finding
an ordering for the coefficient matrix which suffers low
fill-ins when it is factored. Therefore, this scheme
requires a simulation of the effects on the accumulation
of non-zeros of the elimination process. In order to
avoid direct elimination with actual values, a symbolic
factorisation is usually adopted to obtain the zero and
non-zero structures of the factored matrix. As the
numerical values of the matrix components are of no
significance in this connection, the problem could be
studied using a graph approach, instead of using an
actual matrix factorisation. The minimum degree
algorithm is particularly suitable for solving medium to
large networks, e.g. systems with 200 nodes and more.
For smaller networks, the minimum degree algorithm

perturbations, in particular for those systems with

moving block signalling ;

becomes a less efficient option, since a significant

portion of the overall processing time will be used in the
symbolic factorisation process.
3.4

Traction Drive Models

ability to select relevant aspects of simulation to suit

application e.g. to switch on or off signalling and
power supply separately and possibly dynamically;

The basic function of a traction equipment model is to

provide tractive effort (TE) output and current/power
demands according to the given input parameters for the
train movement and power network calculations.

further simplification of input data preparation with

graphical data capture where appropriate or
standardisation of interfaces to permanent way data
bases;

The traction supply voltage with respect to the rolling

stock can vary from -30% to +20% of the nominal value
(IEC standard for DC). A voltage sensitive drive model
is, therefore, essential in achieving accurate electrical
and mechanical representations for the conventional DC
traction equipment. However, for the modern three
phase induction motor drive, the voltage fluctuation on
the train pantograph or collecting shoe is less significant
with the advanced pre-conditioning front-end
technology. The voltage magnitude at the DC link can
remain at a fairly constant level.

further widening of choice regarding output format

e.g. animated run-time monitoring graphics; export
to standard packages such as Excel; export to
systems engineering packages for broader trade-off
studies and systems evaluation requirements;

Two kinds of voltage sensitive drive modelling

algorithm, namely, the detailed [1] [2] and simplified
approaches [19] have been used to represent the traction
equipment. When the required information, such as the
motor terminal characteristics, winding resistance and
reactance, etc. are available, it is often desirable to
model a drive using the detailed approach. However,
the detailed drive modelling approach requires a deep
knowledge of the traction equipment and motor
parameters and these are not always easily available. In
some circumstances such as at a feasibility study or
preliminary engineering stage, it is not always possible
to have enough information and time to model a new
drive comprehensively. The simplified drive modelling
approach based on data fitting and numerical techniques
provides a much easier alternative, which only requires
the high level information such as tractive effort vs. train
speed curves which are generally much easier to obtain.
4.

FUTURE DEVELOPMENT

There can be no doubt that computer modelling and

simulation of railway systems provides an invaluable
tool for engineers to evaluate different what-if
scenarios, which can lead to the minimisation of project
cost and programme overruns. There are a number of
areas envisaged to be worthwhile for future
developments:
more integrated environment for different levels of
simulation, such as the detailed signalling systems
simulator integrated with the traction power network
simulator for studying the power demands and
current transients for re-start situations after

extension to include dynamic passenger flow

modelling and integration with traction performance
calculations to reflect variable passenger loading;
further extension of control aspects to include
control traffic regulation, on-board trajectory
calculation to enable simulator to be used as a testbed for advanced train control concepts;
refinement and validation of software models for
direct use embedded within central control
computers or within train-borne control computer.

5.

CONCLUSIONS

Computer simulation of train performance, signalling

and traction power analysis is long-established in the
engineering design of railway systems. With the
increased speed and improved user interfaces
characteristic of modern computing systems, this trend
will continue with simulation being a routine tool for the
design and evaluation of most aspects of the total
railway system. The trend will be towards a single
integrated package, but with distinct functional
subsystems enabling the user to focus on the aspect of
current concern but without losing the realism of a total
system model and thus preserving knowledge of the
system implications.
In parallel and closely linked to the development of
these off-line design aid programs, the same algorithms
will be incorporated into on-line control software firstly
in an advisory role, but eventually fully embedded into
signalling and control systems. This will require a
carefully structured approach to the question of software
validation and clear thinking about the level of
reliability required from each part of the software.

REFERENCES
1. Goodman, C.J., Mellitt, B. and Rambukwella, N.B.,
1987, CAE for the electrical design of urban rail transit
systems, COMPRAIL 87, 173-193
2. Mellitt, B., Goodman, C.J. and Arthurton, R.I.M.,
1978, Simulator for studying operational and powersupply conditions in rapid-transit railways, Proc. IEE,
125, 4, 298-303
3. Graham, I, 1991, Object oriented methods,
Addision-Wesley Publishing Ltd
4. Taskin, T. and Goodman, C.J., 1991, Modelling of
signalling in an object oriented simulation model,
ASPECT91, pp. 108-128
5. Siu, L.K. and Goodman, C.J., 1992, An objectoriented concept for simulation of railway signalling and
train movements, COMPRAIL 92, vol. 2, 18-20 Aug.,
1992, 545-556
6. Mellitt, B., Goodman, C.J. and Arthurton, R.I.M.,
1978, Simulation studies of energy saving with chopper
control on the Jubilee Line, Proc. IEE, 125, 4, 304-310
7. Gill, D.C. and Goodman, C.J., 1992, Computerbased optimisation techniques for mass transit railway
signalling design, IEE Proc. B, 139, 3, 261-275
8. Kumar, D., 1986, A noval approach to sequential
simulation, IEEE Software, 25-33
9. Cury, J.E., Gomide, F.A.C. and Mendes, M.J., 1980,
A methodology for generation of optimal schedules for
an underground railway system, IEEE Trans. On
Automatic Control, 25, 2, 217-222
10. Van Breusegem, V., Campion, G. and Bastin, G.,
1991, Traffic modelling and state feedback control for
metro lines, IEEE Trans. On Automatic Control, 36, 7,
770-784.

11. Cremonini, R., Lamma, E. and Mella, P., 1988, A

simulator for automatic train protection system, Proc.
of Simulation V Conference, 295-299
12. Ho, T.K., Goodman, C.J., and Norton, J.P., 1995,
An event-based traffic flow model for traffic control at
railway junctions, ASPECT95, 163-170
13. Goodman, C.J. and Siu, L.K., 1994, DC railway
power network solutions by diakoptics, Proc. of 1994
ASME/IEEE Joint Railroad Conference, 103-110.
14. Uher, R.A., 1987, Rail traction energy management
model, COMPRAIL 87,.3, 39-60
15. Talukdar, S.N. and Koo, R.L., 1977, The analysis
of electrified ground transportation network, IEEE
Transaction on PAS, PAS 96, 1, 240-247
16. Digby, G. Hewings, D.B. and Kadhim, R.J., 1992,
Simulation of voltage controlled traction rectifiers on a
medium voltage AC distribution system, COMPRAIL
92, 2, 27-37
17. Cuthill, E. and McKee, J., 1969 Reducing the
bandwidth of sparse symmetric matrices, Proc. 24th
Nat. Conf. Assoc. Comput. Mach, ACM Publ., 157172
18. George, A. and Liu, J.W.H, 1981, Computer
solution of large sparse positive definite systems
Englewood Cliffs: Prentice-Hall
19. Yu, J.G., Struand, R.W. and Denning, L.R., 1994,
A General motor modelling method for transit system
simulation studies, COMPRAIL 94

Branch Up

r 4u'
R '
3u

I 4u'

Branch Down

r 4d '

I 4d'

Up Road
r 1u
R
I 1u
r

I 1d
V1

T E
1u
r 2d

1d
T

I 2d
E

1d

R
I 4u

I 3u

T E
2u
r 3d

2d

I 3d
T

E
2d

r 4d''

r 4d

I 4d
S

R '
3d

r
R3

I 5d'

3d'

r 5u

6d

r 7d'

'

I 6d'

I
T

V3 '

r 6u
I

I 6u
I 5u
T E
T E
3u
4u
r 5d
3d
E

R3
S

3d

E 4d'

I 6d

V3

Down Road

Fig.1 Simplified branched Traction Power Network.

'

7d

V 4'

7u

7u
r 7d

r 6d

I 5d

R 4'

R 4d'

'

R 4u

3u

T
V2

I 7u'

I 6u'
I '
T E 5u T E '
4u
'
3u
r 5d '

R2

r 7u'

r 6u'

R 4u'

r 4u

r 4u''

R 2u

1u

I 2u

1d

R1

r 3u

r 2u

r 5u'

R4

R 4d
T

I 7d
E 4d

S
V4