Equipment Distribution Optimization at BNSF

1
Equipment Distribution Optimization at BNSF
INFORMS Annual Meeting November 2005
2
Equipment Distribution (ED)
3
Equipment Distribution Optimization (EDO)

• EDO

System-Wide, Optimal Solution Solutions
Automatically Revisited
Empties (at next available location) Loads
(future empties at unloading location) Empty cars
expected to be received in interchange All
Supply Defined by Equipment Types (ETs)
Customer Demands Non-Customer Demands
4
Defining SupplyEquipment Typing
5
Defining DemandCustomer Profiles
Customer Location Commodity Request Equip Accept
Window 1 - ET1 1 - ET2 5 - ET3
Preference
Equipment Type Code
6
Trip Plans / PESTs(Projected Empty Spot Times)

• Trip plans are generated for all cars
• Trip plans have ETAs at all points along the way
• Trip plans know when and where cars are planned
  to be switched next
• Trip plans are continually updated
• PESTs are virtual trip plans created from base
  train schedules and used to estimate transit
  times between all historically observed O/D pairs
• Nightly batch jobs update newly observed O/D
  pairs and refresh oldest PESTs

7
Solving

• Identify all possible assignments of supply to
  demand -- equipment match and able to arrive in
  time
• Calculate a value for each possible assignment
• Find the group of assignments that result in the
  greatest overall value
• Re-solve every fifteen minutes -- taking into
  consideration updates to supply and demand since
  the last solve

8
Provide supply and demand updates
Ask for disposition
Give / update disposition
Key
Continuous Process
Sequential Process
Pull supply and demand updates from mainframe
Send plan changes to mainframe
AIX Server
Compare Current and Previous Solutions
Generate Candidates
Solve
9
EDO Rethinks
HAVRE
1
MINOT
SPOKANE
EVERETT,WA
DILWORTH
2
PASCO
NORTHTOWN
3
GALESBURG
KANSAS CITY, KS
STLOUIS
1. Car Released Empty in Everett, WA
Initial Decision is Memphis, TN (with Rethink in
Pasco, WA) 2. Prior to Arrival at Pasco,
triggered for next decision New plan is
Chicago, IL (with Rethink in Galesburg, IL) 3.
Prior to Arrival at Galesburg, triggered for next
decision New plan is Kansas City, KS
(with no Rethinks) Car locked to Kansas
City, KS demand and removed from supply
10
Glint

• Tool to relive past solves
• Able to scrutinize candidates and assignment
  values of past solves
• Used to calibrate solver parameters and test
  what-if scenarios

11
Potential for Shotgun of Assignments
12
Solving the Shuffle Problem
13
Next Steps

• Improve the EDO Assignment Value
• Simplify make it easier to understand and
  calibrate
• Make the Assignment Value as dollar-based as
  possible
• Incorporate Probabilistic Transit Times
• Estimated transit times in EDO are currently
  fixed values derived from base trip plans. A car
  can either get to a destination by a given time
  or it cant there is no maybe.
• Incorporate transit time variability into
  assignment values i.e., discount the value of
  assigning a car to a demand if it has a low
  probability of getting there by the want date.
• Incorporate transit time variability into
  candidate generation i.e., require a minimum
  probability of arriving by the want date for
  various demand types.
• Allow for future evaluation of other solution
  techniques and factors
• Opportunity costs of equipment assign cars to
  demands based on alternate uses in close
  proximity
• If a probable destination of the loaded move is
  known, consider factoring that knowledge into the
  current solution.
• Work with the Ag Commodities group and others to
  address unit train optimization

14
Lessons Learned

• What was done well?
• Implemented a fleet at a time
• Communicated prior to implementation internally
  and externally
• Had knowledgeable people involved
• What would we do differently next time?
• Dont leverage as much of the existing system
• Develop supporting applications before
  implementation
• Agree on measures of goodness upfront
• Build a prototype first then build the production
  system
• Manage expectations especially the need for
  enhancements

15
The end.