You are on page 1of 7

DEALING WITH OPERATIONAL CONSTRAINTS IN

ENERGY EFFICIENT DRIVING


T. Albrecht*, C. Gassel*, A. Binder*, J. van Luipen
*'UHVGHQ8QLYHUVLW\RI7HFKQRORJ\)ULHGULFK/LVW)DFXOW\RI7UDQVSRUWDWLRQDQG7UDIILF6FLHQFHV'UHVGHQ*HUPDQ\
{Thomas.K.Albrecht | Christian.Gassel}@tu-dresden.de, Anne.Binder@mailbox.tu-dresden.de

ProRail, Railway Development, The Netherlands. Jelle.vanLuipen@prorail.nl

Keywords: Energy-optimal train control, Driver Advisory


System, Train Protection Systems.

Abstract
The paper presents algorithms for energy-optimal train
control which can be applied in Driver Advisory Systems. As
the train driver shall be diverted as few as possible by this
system, a constructive two-level algorithm based on the
theory of optimal control and a gradient method is presented
which delivers the energy-optimal regime sequence with the
minimal number of regime changes.
The integration of operational requirements in this algorithm
is discussed in detail, in particular the consideration of the
properties of the Dutch train protection system ATB EG and
the integration of capacity bottlenecks as minimal speed
constraints. The algorithm is applied to some representative
line sections of the Dutch railway network in a case study.

1 Introduction
Driver advisory systems are an appropriate means to reduce
energy consumption of trains [1, 9]. The acceptance of these
systems by the drivers is essential to reach the promised
energy savings in practice. One prerequisite for a high
acceptance is that the computed advice should be consistent
with the safety system in use and the current operational
situation.
Driver advisory systems have first been developed for
regional or suburban trains with relatively short interstation
distances and only simple constraints which have to be
considered. Today, the available processing power and
communication bandwidth allow to integrate even
complicated real-time constraints in the optimisation
algorithm. The paper looks in particular at the Dutch railway
network, where due to the heavily used infrastructure many
complicated constraints occur.
In Section 2 the paper shortly explains the reasons for
choosing a gradient method for the computation of the
energy-optimal trajectory between two consecutive points
with fixed train state. These can either be train stations with
fixed arrival times or capacity bottlenecks with fixed passing
times. The extension of the existing algorithm for the
consideration of capacity bottlenecks is described in Section
3. In Section 4, the Dutch train protection system ATB EG is
analysed in its influence on energy consumption. It is

explained, how the properties of this system can be integrated


in the gradient method to compute trajectories, which are
driveable in the Dutch railway network. Both sections are
concluded with case studies, where the developed algorithm
is applied in some typical examples for regional and intercity
trains. The algorithms have been implemented by Dresden
University of Technology on order of ProRail. They are part
of the software tool BEA which can be used as planning tool
or driver advisory system and which is briefly presented in
Section 5.

2 Energy-efficient driving
2.1 General design questions
Energy-efficient driving between two consecutive stations
has been discussed in the literature since the 1970s (see [1]
for an introduction and [7] for a comprehensive literature
overview).
Two basic approaches can be distinguished:
1. Use of a precise train model and calculation of the
trajectory using a discrete search space and a numerical
solver. The application of Dynamic Programming is the
most popular approach in this field, see e.g. [5], other
search methods like direct search can be found in the
literature as well.
2. Derive the optimal regimes using the Maximum
Principle and construct the trajectory based on these
regimes.
In the first approach constraints for speed and time can be
easily interpreted as restrictions in the search space. The
entire freedom of tractive effort is seen as disadvantage of the
algorithm: the driver can only follow a given advice exactly,
if displays exist which compare requested and actual tractive
effort: This is only the case in a few modern traction vehicles.
The quality and energy-efficiency of such an algorithm
depends on the chosen discretisation of the search space, if
discretisation intervals are large, the method leads to
solutions which are far from optimal in the real world, if the
intervals are small, the tractive effort might change
frequently, which might divert the driver. That was the most
important reason to use the second approach here: By
application of the Maximum Principle one obtains that there
are four optimal regimes, in particular:
x
Full power (limited by the maximal permitted
acceleration of the train, Abbreviation: pow)

station 3

station 2

maximal speed

equations of motion are solved numerically here. A


discretisation interval of 1 s is chosen.
A constructive method is then used to compute the speed
profile for the run on the subsection. First, the timeoptimal
running is computed, which can only contain the regimes full
power, cruising at maximal speed and full braking. It is then
checked, whether the speed on a subsection exceeds the
defined minimal speed. If that is not the case (or only on a
part of the subsection) the part of the subsection where the
speed is below the minimal speed is separated from the
original subsection and is marked as fixed subsection, i.e. it
cannot be modified anymore.
Depending on the available running time for the subsection,
first, a coasting phase and with increasing reserves a cruising
phase below maximal speed are introduced. Coasting takes
place before cruising, if during coasting the speed increases
(on strong descents, depending on the running resistance of
the train and other track resistances), see Fig. 2.
crm

pow

cr

co

only on steep
descents
in all other cases

min speed =
max speed
in bottleneck

speed

station 1

Cruising at constant speed (either by applying partial


power or partial braking, depending on the prevailing
track and running resistance, Abbr.: crm if cruising at
maximal permitted speed, cr in all other cases)
x
Coasting (Inertia motion, Abbr.: co)
x
Braking with full power (limited by the maximal
permitted deceleration of the train, Abbr.: brm)
The existence of these four regimes is known for many years
and also taught to the drivers, who are used to thinking in
these regimes. This approach limits the number of regime
changes and therefore corresponds with the wish of the driver
to minimise control actions during a ride. Drivers are less
deviated from the driving task compared to other frequently
varying advises which is particularly important for safety
on dense and heavily charged railway networks.
An entire train run is decomposed into sections whose start
and end state is fix (t, v, s). This is always the case in stations
(v = 0). In addition, it might as well be the case in bottlenecks
where the speed of the train should equal the maximal speed
although it should be noted that this constraint cannot be
found explicitly in timetables today. Furthermore, a section is
decomposed in subsections with constant track conditions
each (maximal allowed and minimal desired speed, track
resistance). Solutions have to be found for the construction of
an optimal trajectory in two stages: (1) for each subsection of
the track with constant requirements and (2) the transition
states between the consecutive subsections have to be found.
The way of decomposing a train ride into sections and
subsections is illustrated in Fig. 1. In the illustrated example,
track gradients are assumed to be constant for the whole train
ride.

brm

Figure 2: Possible regime sequences for one subsection


(Classical approach without consideration of operational
constraints)

ATB speed
minimal speed
position

subsection 1
section
timetable
requirement

4 5
arrival
time

234

1 2 3

passing
time

arrival
time

Figure 1: Explanation of the problem decomposition


2.2 Determine the optimal regime sequence for a single
subsection
The exact regime sequence and the switching times between
the regimes depend on subsection length, track resistance,
start and end speed, train properties and the running time for
this particular subsection.
Earlier contributions used the synthesis solution of the
linearised equations of motion of the train to determine the
switching points between consecutive regimes. Because
trajectories are required to be as exact as possible over the
entire speed range and because tractive and braking effort of
modern trains are highly non-linear functions of speed, the

2.3 Determine the optimal transition states between all


subsections
It has to be determined how the running time margin which is
available for a section shall be distributed among the different
subsections.
First, the time-optimal driving style is determined on all
subsections. Then it is checked on all subsections, what
energy saving can be achieved by adding a certain amount of
time margin to that particular subsection. If it is not
possible to add more running time to a particular subsection
with the current start and end speed, it has to be checked
whether and how much energy can be saved by reducing the
VXEVHFWLRQV HQG VSHHG by an amount . Any restrictions
(e.g. due to minimal speeds) must be considered.
For each of the subsections i one of the alternatives to save
energy is simulated. The amount of energy saved with the
alternative driving style is divided by the change in running
time (on the whole section) caused by this modification.

This so-called gradient


=

old alt,
alt, old

is compared for all subsections and all variants for energysaving, the maximum is determined and the corresponding
alternative transition states are set. The algorithm is repeated
until all running time margin has been distributed or it is not
possible to add more reserve to any of the subsections which
might be the case if minimal speeds would be hurt by
applying more margin.
Of course the step sizes and determine the precision of
timekeeping and energy efficiency. They are adapted
depending on the amount of running time margin which still
has to be distributed.
This gradient method leads to optimal solutions when only
small running time reserves are available. The more margin
has to be distributed, the bigger the chances are, that the
stepwise distribution of the margin only leads to suboptimal
solutions. The possible consequences will be discussed in the
case studies.

3 Energy-efficient driving under consideration


of capacity bottlenecks
3.1 Necessary adaptations to the optimisation algorithm
In order not to get hindered by a preceding train, a train shall
not pass a location earlier than a cHUWDLQWLPH HDUOLHVWWLPHconstraint). In order not to hinder a following train, a train
has to be at a certain location not later than a given time
ODWHVWWLPH- constraint)1.
In dense networks, these restrictions might follow shortly one
after the other which corresponds to a so-called fixed passing
time. In these situations, the required average speed between
the restrictions is given by the timetable. In bottlenecks this
required speed almost always corresponds to the maximal
speed. Examples of practical importance are situations where
one train overtakes another or a dense sequence of trains
entering a station e.g. in synchronised timetables.
To deal with these constraints, new regime sequences are
needed which have to be applied to prolong running time on
subsections with fixed (high) end speed. Normally, on
subsections with positive running resistance the slowest ride
to reach a defined end speed from a smaller start speed in an
energy-efficient way (see 2.2) consists of accelerating from
start speed to end speed and then cruising until the end of the
subsection with the required end speed. However there might
be situations, where this time is still too short in order to
fulfil the operational requirements, e.g. if the section end
speed corresponds to the maximal speed. As the end speed of
the subsection is fixed, the total length and duration of the
1

,QWKHRQJRLQJUHVHDUFKSURMHFW5DLOHQHUJ\WKHVHWLPHV
are referred to as restrictive target point. They are part of the
proposal for a communication standard between operation
control centres and onboard units [10].

acceleration process on the subsection is fixed as well. The


only possibility for prolonging running time and energy
saving consists of modifying the cruising phase, i.e. applying
a sequence of 1st Acceleration 1st Cruising at speed 1 2nd
Acceleration (to subsection end speed) 2nd Cruising (at
subsection end speed 2 ). The speed to choose during the
first cruising phase depends on the duration of cruising at the
end speed.
ss acc 2 2
1 =
ss acc
where ss is the length of the subsection, ss the available
running time for the subsection, acc and acc the time and
length needed for acceleration to the end speed and 2 the
time of the final cruising phase on the subsection.
Total energy consumption during both cruising phases must
become minimal. It can be computed using the simple
equation
1
1
=
+
1 res 1 1 2 res 2 2
where res is the quadratic function representing the
driving resistance of the train at a certain speed and the
traction efficiency at that speed. As long as traction
efficiency at 2 is not significantly higher than at 1 , it can
easily be shown that it is energy-optimal to cruise as long as
possible at the slower speed 1 and not to cruise at subsection
end speed at all.
Other combinations of start, end and minimal speed
constraints can be dealt with in a similar way. This finally
leads to the definition of the following new regime
sequences:
x full power cruising full power
x braking cruising acceleration (the so-called bathtub-curve, see [6])
x braking coasting (cruising) braking
These sequences must be applied, if either minimal speeds or
time constraints would be hurt by applying only the energyefficient regime sequences as described in 2.2. Note that on
strong descents, cruising can be replaced by coasting
regimes.
If it is not sufficient to introduce the above mentioned
regimes on one subsection, but on several consecutive
subsections with different gradients, the height profile should
be exploited wherever possible. It might e.g. be better to use
a second cruising phase, if that takes place on a downhill
section. A general approach to consider this is currently
under development.
3.2 Case study
The consideration of dynamic operational constraints shall be
discussed on the example of the line Utrecht Den Bosch,
which is 48 km long and one of the busiest lines in the
Netherlands. The most critical bottleneck on the line is the
station of Geldermalsen approximately in the middle of the
line, where Intercity trains regularly have to overtake regional
trains. This is an example of an exact passing time definition
or mandatory target point [10]: If the Intercity is too early, it
will be hindered by the regional train in front, if it drives

140

Speed /km/h

120
100

= 960s

Gdm

Earliest passing
time in Gdm

150

Energy

110

100

Delay
in Ht

100

Delay in Ht /s

120

50

850

900

950
1000 1050
Passing time /s

1100

1150

Figure 4: Dependency of energy consumption for the whole


trip on the passing time in the overtaking station

4 Static Influence of the Dutch ATP system on


energy-efficient driving
Modern continuous ATP systems like ETCS Level 2 or LZB
make use of a continuous (radio) communication channel
from trackside to train and an onboard positioning module.
The latter continuously computes the position of the train so
that the distance to the next danger point is known onboard
and safe operation of the train can be supervised by
comparing the braking curve needed to slow down to a
certain point with the actual speed of the train and intervene
in case of non-respect. These systems are in operation in
Europe on high-speed lines or lines with high traffic demand.
However, most of the existing railway networks are equipped
with intermittent train control systems. These systems are not
able to determine the distance to the next danger point.
Instead rules are defined, which have to be followed in order
to avoid critical situation. The influence of the limited
amount of information on energy-efficient driving shall be
illustrated by means of the Dutch ATP system ATB2
(Automatische Trein Benvloeding), which is in operation on
the conventional railway network on lines with line speeds
above 100 km/h.

80
t

60

Gdm

= 1400s

energy-optimal
(858s)

20
0
50
25
0

4.1 Properties of the Dutch ATP System ATB

Gdm

40

Height /m

Energy consumption /%

slower than maximal speed, it will hinder the regional train in


its departure process and therefore cause a delay.
Fig. 3 shows different trajectories on the track. Two time
constraints shall be respected: a (dynamic) passing time at
Geldermalsen (Gdm) and the planned arrival time in Den
Bosch (Ht). The effects of the ATP system have been ignored
in this example. In addition, it is assumed, that the train is
diverted exceptionally to another track and therefore the
speed limit in Geldermalsen station is only 80 km/h.
The continuous black line shows the unrestricted energyoptimal trajectory of the train where Geldermalsen is passed
858 s after departure in Utrecht. It can be seen that the height
profile is exploited during cruising at maximal speed (bridge
at km 17). The dashed line gives an example for a delayed
passing time. Here, cruising below maximal speed is applied
in order not to reach Geldermalsen before time: The next
time constraint can also be fulfilled (arrival time at Den
Bosch). The dotted line shows an example, where due to the
constraints one of the newly introduced regime sequences is
applied (acceleration cruising acceleration). Note that for
path including the new regime sequence the exploitation of
the height profile is not yet implemented.
The energy consumption for the whole ride depends on the
exact passing time as can be seen in Fig. 4. If the passing
time is later than 980 s, the planned arrival time in Den Bosch
(2nd time constraint) cannot be reached, which explains why
total energy consumption decreases again. In real operation
the first timetable restriction will always be the most
important, because the driving on all consecutive sections
depends on when and how the first passing point is passed in
reality.

10000

20000
30000
Position/ m

40000

Figure 3: Comparison of optimal rides for three different


passing times in Geldermalsen (Gdm)

The Dutch system ATB has been installed on all major lines
in the network since the 1950s. It is based on a continuous
communication from track to train, which is realized by
stamping the 75 Hz signal of the track circuits with a
rectangular signal of different frequency the so called
coding of the track circuits. Five different codes can be
transmitted this way from track to train, each code stands for
a certain speed (40, 60, 80, 130 or 140 km/h) which shall be
referred to as ATB speed in the suite of this paper. The ATB
speed is the minimum of the current line speed, the
announced next speed limit and the speed limit announced at
the previous signal. If this maximal speed is not contained in
2

There are two different ATB systems installed. EG stands


for first generation and NG for new generation. The latter is
close to ETCS Level 1 in its functionality but only applied on
regional lines in the Netherlands. The ATB EG has a
significantly higher importance and will therefore be
analysed here and simply be called ATB.

the set of ATB speeds, the next higher value is transmitted


(e.g. at a maximal speed of 70 km/the ATB speed 80 km/h is
sent to the train).
The onboard unit of the ATB system compares the current
speed of the train with the current ATB speed as received
from track side. If the train speed is higher than ATB speed,
the so-called braking criterion has to be fulfilled at least 2
out of 7 braking steps have to be applied otherwise the train
will be automatically brought to a standstill. This system is
not able to observe whether the achieved braking rate is
sufficient to reach the next speed limit, because it does not
have information about the position where this next speed
limit starts.
Another particularity of this system occurs in the so-called
stop door-situation on regional train lines. Here, level
crossings are often situated immediately behind the stations.
The level crossing is closed only when the train has stopped
in the station in order to hinder road traffic as few as possible.
This also means, that while the train is approaching the
VWDWLRQWKHVWDWLRQVH[LWVLJQDOVKRws red (because the level
crossing is not yet closed) and the lowest ATB speed (40
km/h) is therefore sent to the train during the entire approach.

smaller than ATB speed. Then, the optimal regime


sequence consists of ATB braking, coasting and maximal
braking. If strong downhill slopes are situated on the
track, coasting would lead to an increase in speed and
must be replaced by cruising at ATB speed.
- The end speed of the subsection is reduced by the
gradient method. This is only efficient, if either the next
subsection is downhill or coasting can be applied there.
Then the optimal regime sequence consists of ATB
braking, cruising at ATB speed and coasting until the
subsection end speed.
Possible sequences of regimes on a section where driving is
influenced by an ATB speed restriction are given in Fig. 5. If
the subsection entrance speed is below the ATB speed, the
maximal speed of the subsection is set to ATB speed and the
subsection is treated as ordinary subsection like described in
Subsection 2.2 of this paper.
br
ATB

cr
ATB

4.2 Integrating the requirements in the regime sequence


According to the theory of optimal control, braking with
maximal allowed braking rate has to be applied in order to
arrive at the next target point, be it a station or a speed
restriction. The application of a high deceleration rate
guarantees, that the least time is lost during braking (This
time shall rather be used to switch off traction earlier).
With the ATB system in use, it is not possible to brake as late
as necessary, but braking has to start immediately after the
speed reduction is announced. This clearly interferes with the
energy-optimal sequence of driving regimes.
Therefore it has to be found, which braking style is the
energy-optimal one under the given constraints. The optimal
braking strategy is the one where the least time is needed.
Time can be computed with

co

brm

Figure 5: Possible regime sequences on an ATB subsection

140

Timeoptimal
with ATBl

120

i.e. that for a given start state of the train at the beginning of
the subsection (1 , 1 ) and a given target state of the train at
the end of it (2 , 2 ) the trajectory )( = with maximal
area under the curve is the one with the shortest time. It can
easily be shown that this is the curve which consists of
braking with the minimal required number of braking steps
(as long as possible) before either cruising at ATB speed or
braking with maximal braking rate (until the end speed of the
subsection).
In the optimisation process, a change of the ATB speed
restriction is another criterion to form a new subsection. At
the start of the gradient method, ATB subsections are
simulated with time-optimal riding first. During the gradient
method, two situations might occur:
- Running time reserve is added to the ATB subsection:
This is only possible, if the subsection end speed is

100

Speed /km/h

= 2 = 2 ,

80

Energyoptimal
without ATB

60
40

Energyoptimal
with ATBl
Subsection with
ATB speed 40 km/h

500

1000

1500

2000 2500
Position/ m

3000

3500

4000

Figure 6: Simulation of different runs on a "stop-door"


section of a regional train with and without ATB in use
4.3 Case study
To examine the consequences of the ATB system on energy
consumption, several cases have been studied in [3]. This
paper summarizes the main effects for two typical scenarios
for the Dutch railway infrastructure.

A run of a regional train was simulated on the section


Harderwijk - Ermelo using MAT64 train characteristics. On
this 4.4 km long basically flat section, the last signal before
Ermelo station is already situated 1.2 km in front of the
station. This signal will typically show the yellow aspect
EHFDXVH WKH VWDWLRQV H[LW VLJQDO LV GHSHQGLQJ RQ WKH FORVXUH
of the level crossing. That means that the train will get an
ATB speed signal of 40 km/h for the rest of its ride. In Fig. 6
both the time-optimal ride and the energy-optimal trajectory
EDVHG RQ WKH FXUUHQW WLPHWDEOH  UXQQLQJ WLPH  DUH
plotted. In the ATB subsection, both trajectories consist of
the regimes ATB braking and maximal braking.
The same section has been simulated assuming a modern
train protection system like ETCS level 1 or 2, where the real
braking curve of the train is observed. It can be seen from
Fig. 6, that the same running time can be achieved with a
coasting speed of 101 km/h compared to 108 km/h with
ATB. About 20% energy could therefore be saved on this
particular section. Applied to an entire line, an average saving
of 5% can be obtained by replacing ATB with state of the art
train protection systems [3].
An Intercity run from The Hague HS to Rotterdam has been
simulated to show the effect of ATB with different static
speed restrictions. These are of particular importance for long
GLVWDQFHWUDLQV7KHSODQQHGUXQQLQJWLPHRIDULGHLV
A loco hauled train is simulated (E1800 electric locomotive
with 7 Intercity coaches). Fig. 7 shows the application of the
regime sequence ATB braking maximal braking and ATB
braking cruising during the time-optimal run (at kms 6 and
15 respectively). During the energy-optimal run, the gradient
method determines the transition states between the
subsections in such a way, that ATB braking is practically
avoided. The running time margin is almost 15% in this case,
this explains the flexibility which exists in the gradient
method and which leads to solutions, which result practically
in the same energy consumption (difference smaller than
1%). For smaller running time margins (3%), the energy

Speed /km/h

Timeoptimal
with ATB

100
Energyoptimal
without ATB

Energyoptimal
with ATB

50

Height /m

5 Field tests and simulations with the advisory


system BEA Online
The developed algorithms form the core of a software suite
called BEA (Basic Energy Analysis) which can be used
offline to estimate energy consumption and online in driving
simulations as well as onboard of real trains as driver
advisory system, see Fig. 8.
BEA Online today mainly serves as testbed for ManMachine-Interfaces. Drivers are given different displays and/
or acoustical information like:
optimal regime (e.g. the beginning of coasting indicated with
a blue point or acoustically)
comparison of optimal and real speed profile in diagram
plots (all relations between time, speed and position can
be displayed)
speed advice as number or pointer in an analogue meter
deviation from optimal path and tendency of deviation
change
First experiments took place in 2009 within the simulator
MATRICS [2,8] as well as onboard real trains. The
functionality of the tool and the algorithms has been proven
with a relatively simple prototype, which has position, time
and speed information as only real-time input. Other
information (like delay of trains in front) had to be entered
manually using the developed user interface.
Drivers understood how changes to their timetables could be
smoothed away with a speed advice up to the next restrictive
target point. However, this usually means actively braking to
a lower speed. Without BEA Online, drivers would only have
coasted, even though in theory braking is optimal here
because the low target speed cannot be reached with coasting
alone.

6 Conclusions

150

0
20
0
-20

consumption is 7% lower with a modern train control system


compared to ATB.

5000

10000
Position /m

15000

20000

Figure 7: Simulation of different runs of an Intercity train


with and without ATB in use

The paper presents algorithms which can be used for the


computation of energy-optimal train paths. It was shown that
all problems can easily be dealt with separately, but the
combination of the problems is hard to solve. The proposed
decomposition approach delivers a suboptimal solution with
a small number of regime changes, which can easily be
implemented by the drivers.
The Dutch infrastructure manager ProRail expects to increase
capacity by applying advisory tools like BEA Online which
make the trains drive closer to their planned path. If energyefficiency is considered in the computation of this path, train
operating companies can benefit from this strategy as well as
society as a whole.
It still has to be examined how the developed algorithms and
systems can be integrated in a more global framework for
train dispatching (see e.g. [4]) and control to further increase
the impact of such a system. Other interesting points of
research include the definition of a minimal time a regime
should be applied in order to obtain trajectories which can
easier be implemented by the drivers.

References
[1] 7$OEUHFKW(QHUJ\-HIILFLHQWWUDLQRSHUDWLRQIn: I. A.
Hansen, J. Pachl. (eds.) Railway Timetable & Traffic.
Eurailpress, pp. 83-105, (2008).
[2] T. Albrecht, J. van Luipen. :KDW UROH FDQ D GULYHU
LQIRUPDWLRQV\VWHPSOD\LQFRQIOLFWVLWXDWLRQV",Q 10th
IFAC Symposium Control in Transportation Systems,
Delft.. (2006)
[3] A. Binder, Integration of the properties of the Dutch
train protection system ATB EG in Driver Advice
Systems for Eco-driving. Studienarbeit, Dresden
University of Technology (2009).
[4] $ '$ULDQR 7 $OEUHFKW 5XQQLQJ WLPH UHoptimization during real-WLPH WLPHWDEOH SHUWXUEDWLRQV
In: Computers in Railways X, pp. 531540, WIT Press,
Southampton.
[5] R. Franke, R., M. Meyer, P. Terwiesch. Optimal
Control
of
the
Driving
of
Trains.
atautomatisierungstechnik. 50, pp. 606-613. (2002)
[6] M. Luethi. Improving the Efficiency of Heavily Used
Railway Networls through Integrated Real-Time
Rescheduling. PhD Thesis. ETH Zurich (2009)
[7] 5/LX,0*RORYLWFKHU(QHUJ\-efficient operation of
UDLO YHKLFOHV Transportation Research Part A 37, pp.
917-932, (2003)

[8] J. van Luipen, 60HLMHU8SORDGLQJWRWKH0$75,&6


Combining simulation and serious gaming in railway
VLPXODWRUV ,Q 3rd European Conference on Rail
Human Factors (ECRHF), Lille. (2009)
[9] I. Mitchell. The Sustainable Railway: Use of Advisory
Systems for Energy Savings. IRSE News, no. 151, pp.
2-7 (2009)
[10] Railenergy project website.
Address: http://www.railenergy.org (Access: 4.1.2010)

Figure 8: User interface and current usage of BEA Online tool

You might also like