07.03.05 07002-irse-bris2007-rail sched-finaldrc
DESCRIPTION
A presentation accompanying a paper on the subject of railway schedule optimisation and infrastructure optioneering.Presented before the Institution Of Railway Engineers in 2007 by author David CaldwellTRANSCRIPT
IRSE 2007 Technical Conference- Brisbane
Application of Problem Space Search to Heavy Haul SchedulingAlex WARDROP & David CALDWELL
04/12/232
What is scheduling?
“Scheduling” is a broad term and can include crew rostering,
maintenance scheduling and train planning (timetabling)
Here the main interest is train planning
Train plans may be developed around the constraints of crew rostering,
and in cooperation with maintenance scheduling
04/12/233
Train Planning
Train planning is placing a set of train requirements over a railway
network
Train requirements are the trips that train operators want to operate
over a given period of time to meet commercial requirements (moving
stuff)
The aim of the train planner is to move these trains to their required
destinations with as little delay as possible
04/12/234
Train Planning
Train planning Maintenance scheduling
Crew rostering/ industrial requirements
Train demand (e.g. trains, origins, stops,
destinations)Maintenance requirements
Day of Operation planning
Operations
Maintenance
Other Constraints
Network (e.g. running times, loop lengths, safeworking,
loader capacity) Train performance and requirements
04/12/235
Developing a timetable
In a manual process, the highest priority trains are “drawn” on the plan
first
Lower priority trains are stopped for passes in preference to stopping
high priority trains
• Order of priority- “two legs, four legs, no legs”
This means that passenger trains go through the route on the minimum
running times
04/12/236
The heavy-haul problem
While the most densely utilised heavy-haul lines are double track, the
majority of Australia’s heavy-haul track mileage is on single-track line
A major challenge is figuring out how to stop and pass trains, and when
to admit them to the next section This has to be done with consideration for
both safeworking requirements and
refuging opportunities
A train plan looks like this, showing the
train ID and stops
Lines are coloured to represent an
important attribute, like train length,
running time, or priority Dist
ance
Time
04/12/237
What is Problem Space Search?
Unlike a mathematical algorithm which seeks to find an exact solution,
Problem Space Search employs a heuristic (or trial and error
mathematical process) to develop many feasible solutions to a problem
These solutions can then be scored against an objective function and
ranked
Research undertaken jointly by the University of South Australia and
the former TMG International found that the technique was applicable to
rail scheduling
04/12/238
Application of Problem Space Search
By applying perturbations to crossing delays of trains moving through a
network, the order in which trains meet is varied, and so the orders of
crossings changed
Thousands of different valid train plans can be generated in a matter of
seconds
These valid solutions can then be evaluated and ranked for their
performance against criteria such as delays and cost
04/12/239
Application of Problem Space Search
Developing a timetable by manual techniques generally takes months
An automated train plan optimising technique has obvious direct
benefits for train planning, reducing the time to produce a timetable,
and improving its “quality”
The huge reduction in the effort required to develop a timetable means
that timetables can be quickly manipulated very close to the
implementation horizon, and still produce optimal results
Timetables can also be easily developed for speculative purposes
(“what if…” scenarios)
Software using Problem Space Search for these purposes has been
developed
04/12/2310
Line Capacity report
Near-optimal timetable
RailSched
Information flow
RAIL NETWORK DESCRIPTION
Track layoutRefuge locationsJunction locationsSectional running timesSafeworking systems
TRAIN CHARACTERISTICSLengthRunning time classPriority INPUT TRAIN REQUIREMENTS
Nominal despatch timesDays of operationDwell timesDependencies between services
04/12/2311
Typical Inputs
Track Infrastructure- links and nodes
Bi-directional line
Uni-directional line
Refuge
Crossing loop
Crossovers
Station name
04/12/2312
Output: optimised train plan
BW4 is a high priority train
MB3 is a low priority train
04/12/2313
Output: Capacity utilisation
04/12/2314
Application
The Problem Space Search technique is applied for both timetabling
and strategic infrastructure planning
An integrated scheduling system, ScheduleMiser ™ was developed for
BHP Billiton
A more general scheduling tool, Rail//Sched ™ is the subject of
ongoing development for the Australian Rail Track Corporation
In both cases the general data requirements and outputs are the same,
thought the interfaces differ significantly, according to specific
requirements
04/12/2315
Case study: ScheduleMiser
ScheduleMiser was developed to represent BHP Billition’s iron ore
moving process, rather than looking at trains in isolation
Time it takes to load a rake
Time it takes to re-stack the stockpile at the loader
Desirability of different rake configurations
Material at mine loader
04/12/2316
Translating BHP requirements
The user specifies production requirements, as well as other
operational variables like loader availability and rake locations
ScheduleMiser builds trains of one, two or three rakes to satisfy
production
Day to day operational conditions, like TSRs and track possessions, are
also configured
04/12/2317
The BHP requirement
Rolling stock, mine, unloader and network
configuration
Ore requirements at Shipping port
Optimised timetables and train plan
Rake reportDumper report
ScheduleMiser
Track availability (possessions)
04/12/2318
Typical schedule output
The user configures all the requirements and runs the schedule
resolver
A train plan, similar to the following, is resolved
04/12/2319
Planning close to the horizon
Need to be able to rapidly review timetable for changing train
requirements, e.g. out of order running
Developing amended timetables close to the event
WorleyParsons, with the assistance of a Commercial Ready grant, is
currently integrating automated scheduling with real-time GPS train
location information
Train locations can be continuously updated in the model, and a
timetable with minimum delay calculated almost immediately
There is another significant application of automatic timetable
generation…
04/12/2320
Strategic Infrastructure Planning
Why different scheduling scenarios need to be considered
• Which loops should be lengthened or moved?
• Will changing the safeworking system (i.e. eliminating token
exchange delays) have a practical effect on capacity?
• Will automating points (and reducing route-setting delays) and
improving turn-out speeds have a practical effect?
• What is the effect of running some longer trains which are excluded
from more loops?
04/12/2321
Considering the future
DEMAND SCENARIOS
SCHEDULING
WHAT IS THE MOST COST - EFFECTIVE?
INFRA - STRUCTRE
TRAINS
OPERATIONS SCENARIOS
DEMAND SCENARIOS
SCHEDULING
WHAT IS THE MOST COST - EFFECTIVE?
INFRA - STRUCTRE
TRAINS
OPERATIONS SCENARIOS
04/12/2322
Case Study: Rail//Sched
The ARTC is developing its network for expected traffic growth
Their 2005 Hunter Valley strategy anticipates that from 2006 to 2010,
tonnage between Musswellbrook and Antiene will approximately triple
from about 20MTpa to about 60MTpa
04/12/2323
Hunter Valley growth
From
ART
C 20
05 H
unte
r Val
ley
Coal
Net
work
Cap
acity
Impr
ovem
ent
Stra
tegy
, fig
ure
A, p
ii
04/12/2324
The effect of growth
ARTC developed infrastructure improvement plans to satisfy this growth
To verify that delay would remain at an acceptable level, ARTC
developed proposed infrastructure schemes (varying over time), and
then developed experimental future train requirements to meet
operators’ train requirements/ tonnages
ARTC then used Rail//Sched to generate hypothetical timetables for the
increased network usage
Once these timetables were generated, they were assessed for relative
changes in delay compared with current operations
04/12/2325
ARTC’s application
Expectedly, when traffic
levels increase, but the
network remains
unchanged, average
delay per train and travel
times increase
The process applied to
assess infrastructure
improvements…
Describe infrastructure scenario (s)
Develop input timetables (train types and despatch times of projected traffic)
Optimised timetables Average Delays
Post processing (Excel)
RAIL//SCHED
Pre-process (add noise to despatch times)
Normalised dimensionless delay
04/12/2326
ARTC’s application
The types of network variables that are typically considered are
safeworking, loop lengths, loop locations and additional tracks
One component of ARTC’s strategy to meet projected demand is the
Antiene to Grasstree duplication
Duplication by 2008, proposed in the strategy, reduces delays to below
the levels in early 2006
04/12/2327
Modelling delay
Modelled Hunter Valley Network Performance
KIY M
DX B
NX
BE
L
MB
H
MB
1
WH
M
SG
L
CA
M
MO
C AC
S
NJM
ND
C
DR
JA
NT
GT
R SL
R
MU
A
0
5
10
15
20
25
30
35
Ex
pe
cte
d d
ela
y p
er
tra
in
(dim
en
sio
nle
ss
tim
e)
4th quarter 2006
1st quarter 2007
2nd quarter 2007
3rd quarter 2007
4th quarter 2007
1st quarter 2008
2nd quarter 2008
NEWCASTLE MUSWELLBROOK
Delay per train reducing in first half of 2008
Duplication of Antiene - Grasstree in first half of 2008
04/12/2328
Benefit for planning
Being able to quickly generate these speculative train plans (and
calculate delay) is very helpful for considering
• Whether a proposed infrastructure plan is a going to satisfy
requirements
• Which projects should take priority
• Whether changing traffic demands will adversely effect delay (and
operators’ costs)
04/12/2329
Conclusion
Scheduling is at the core of railway operations
Train plans which reduce the amount of delay per train improve the
capacity and efficiency of the railway
Optimisation of train plans by application of the problem space search
technique is much faster than manual techniques
Computer generation of train plans makes it possible to experiment with
effects of train and infrastructure changes, and assess the usefulness
of capital projects