The Little strech from Airline Tail Assignment to Railway Fleet Routing

1
Conducting a tight benchmark and beating clients
expectations!
Sami Gabteni, Ph.D. Carmen Systems, Railway
Solutions
Informs-RASIG, San Francisco, November 13th-16th
2005
2

• The benchmark
• European process Fleet Assignment and Fleet
  Routing
• Choosing the approach
• Modelling issues
• Tools at hand Rail Fleet Assignment vs Air Fleet
  Routing
• The Fleet Routing algorithm
• Implementation work
• Adjustment areas
• RAVE
• Examples
• Results
• CRS Fleet and the DSB project

3

• Context
• Complete train units
• IT replacement study
• Too much manual processing
• Client focused on feasibility workshop capacity
• Evaluation based on troublesome Fleet Routing
• We also wanted to demonstrate savings

• Challenge
• 1,5 man/month 3 calendar weeks
• Excludes
• Setting the problem understanding, scope,
  evaluation frame
• Time for client to provide the required data
• Includes
• Modelling
• Data interfaces
• Tightening specifications, modelling, solution
  approach
• Some feedback loops with the expert planners
• Creating a few specific reports

4
Assignment Value (vehicle type, task)
Constraints
More detailed constraints
5

• Restrictions
• 1 or 2 units per composition
• Deadheads (DH) to through workshops
• Minor maintenance services with inclusion
  relations
• Workshop capacity constraints
• Variable connection times depending on shunting
  operations

• Objectives
• maintenance services
• DH kms
• Shunting operations

• Problem size
• Clients biggest problem 30 units, 1900
  trains, barely an issue
• Our biggest problem 70 units, 4000 trains,
  required some tuning

6
Airline Fleet Routing Tail Assignment
Railway Fleet Assignment

• Multi-commodity network
• flow problem solved
• by Lagrangean Relaxation
• Sub-gradient

• Set partitioning problem
• solved by DW-decomposition
• RCSPP
• Heuristic branching
• CP based look ahead

Railway Fleet Routing

• One commodity per unit
• Resource constraints

• Shunting operations

• Representation of compositions
• Daily vs Cyclic

7
Adjustment areas
Business modelling issues
Effective implementation work

• Generalize the concept of task

• 2 units per composition
• Maintenance workshop constraints

• Connexion legality definition

• Deadheads (DH) to through workshops
• Minimum connection times depending on shunting
  operations

• Minor maintenance services with inclusion
  relations

• Extension to the resource
• constraint definition

• Objective adjustments are straight forward

8
Adjustment areas
Fleet Routing Algorithm
RAVE -Configuration Module
Business rule Modelling
Minimum connection time Shunting operation
complexity Station to maintenance DH
Feasible connection

• Resource accumulates
• Updates
• Resets

days w/o break inspection DH kms

• Focus on configuration
• Architecture made it possible
• Reduced project risk

9

• Huge test sets with very diverse structures
• Exact modelling approach
• RD and Consultants to focus on their respective
  issues
• Common repository for business rules
• User control

10
is_connection_legal is_same_station_co
nnection (AND is_allowed_manouvering)
AND is_min_connection_time_legal .... is_min_
connection_time_legal connection_time
gt min_connection_time connection_time
next(leg(chain), departure) - arrival min_conne
ction_time parameter 0030 / simplification
/ remark standard minimum connection
time is_same_station_connection
arrival_station next(leg(chain),
departure_station)
11
Reverse Station
Through Station
is_allowed_manouvering let same_dir
cxn_mod_number next(leg(chain),
cxn_mod_number), pos_in
trainUnit_position, pos_out
next(leg(chain), trainUnit_position)
AllowedManouvering(pos_in, pos_out, same_dir,
station_topo) ... Manouvering_time(pos_in,
pos_out, change_dir, station)
12

• All problems solved in reasonable time
• Minimum implementation time
• Scalability Large mixed problem
• 4 CPU min. first solution
• 2,5 CPU hours to convergence

• Introduced
• Maintenance workshop capacity control
• Re-positionnings through maintenance

• Minor maintenance services reduced by 20
• Shuntings reduced by 30
• Deadheading to maintenance by 35 (fuel, slots,
  utilization)

13
Ongoing DSB development
Fleet Assignment With global constraints Or
Composition Assignment
Restrictions on composition transitions
Composition Assignment With side constraints
14
Ongoing DSB development
15

• Rail domain knowledge
• Understand the issues
• Quickly decide on how to go about them
• Creativity on the tools side
• Business rule encapsulation
• Focus on problem formalization
• Reduce project risk
• Established process with existing decision
  support
• Faster to scope
• No data issue

sami.gabteni_at_carmensystems.com