Title: GECCO 2001
1Experience from designing transport scheduling
algorithms
Raymond Kwan School of Computing, University of
LeedsR.S.Kwan _at_ leeds.ac.uk
Open Issues in Grid Scheduling Workshop, Oct
21-22, 03
2Outline
- Public transport scheduling
3Public transport service
4Planning and scheduling
- Operator one optimisation problem, all decisions
are variables
- Some decisions are fixed by earlier tasks
- Some decisions are left open for later tasks
5Planning and scheduling tasks
Service and Timetable Planning
Vehicle Scheduling
Crew Scheduling
Crew Rostering
6Research Development at Leeds
- Span over 40 years (22 years myself)
- Algorithmic approaches- hueristics- integer
linear programming- rule-based/knowledge-based-
evolutionary algorithms- tabu search- constraint
based methods- ant colony
- Numerous users in the UK bus and train industries
7Parties involved in UK train timetabling
8Train timetables generation
- Three key types of decision variable
- Resource options at a station
9Hard Constraints
- Headway time gap between trains on the same track
- Junction Margins time gap between trains at a
track crossing point
- On a track
- At a platform
10Soft constraints
- (TOCs) Commercial Objectives
- Preferred departure/arrival times
- Efficient train units schedule
11Bus Vehicle Scheduling
- Selection and sequencing of trips to be covered
by each bus - Each link may incur idling or deadrun time
- Minimise fleet size, idling time, deadrun time
- Other objectives e.g. preferred block size,
route mixing
12Bus Vehicle Scheduling - FIFO, FILO
13Driver Scheduling - Vehicle work to be covered
( Relief opportunity )
142-spell driver shift example
Vehicle 1
Vehicle 2
Vehicle 3
15More example potential shifts
Vehicle 1
Vehicle 2
Vehicle 3
16Some characteristics of vehicle and driver
scheduling
- Jobs to be scheduled have precise starting and
ending clock times
- Scheduling involves trying to get subsets of jobs
to fit within their timings to be collectively
served by a resource (vehicle or driver)
- Not the type of problem where jobs are queued to
be served by a designated resource
17Driver Rostering
- To compile work packages for driverse.g. A
one-week rota
- Rules on weekly rotas
- Drivers may take the rotas in rotation
- Optimise fairness across the packages subject to
rules and standby requirements
18Multi-objectives what is optimality?
- Operators do not always try equally hard to
achieve optimal operational efficiency
- Problem at hand is not on the critical path
19Global optimisation?
- Automatic global optimisation is obviously
impractical
- Combining two successive tasks for optimisation
are sometimes desirable, e.g.
- Hong Kong fixed size fleet, fixed peak time
requirements, schedule buses maximise off-peak
service
- Sao Paolo driver and vehicle tied schedules
- First (UK bus) ferry bus problems
20Better optimisation through intelligent
integration of the scheduling tasks
- Sometimes superior results could be simply
obtained where powerful optimisation algorithms
fail
- A more favourable scheduling condition could be
achieved from the preceding scheduling task
- E.g. driver forced to take a break after a short
work spell swap in the vehicle schedule to
lengthen the work spell
- Needs good vision from the human scheduler
rule-based expert system to integrate the
scheduling tasks?
21Scheduling for different service types
- Different types of service may pose different
levels of difficulty for scheduling (different
algorithmic approaches?)
- Urban commuting high frequency, many stops
- Sub-urban and rural lower frequency, fewer stops
- Inter-city and provincial long distance, few
stops
- Some problems have to consider route and vehicle
type compatibility
22Discussion