IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
(PRESENTED AT INFORMS - MAY 8, 2000)
- HISTORY OF LDM (LOCOMOTIVE DISTRIBUTION MODEL)
- DATA ASPECTS
- PROCESS ASPECTS
- LESSONS LEARNED
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
HISTORY OF LDM DEVELOPMENT & IMPLEMENTATION
- STRATEGIC MODEL (1995 - 1998)
STEADY STATE
SATISFY TRAIN REQUIREMENTS AT MIN COST
USED FOR FLEET MANAGEMENT, TRAIN PLAN EVALUATION
- TACTICAL MODEL (1997 - )
1997 AUTOMATED DATA FEEDS (TRAINS, MAINTENANCE)
TONNAGE FORECASTING (DoW)
CLASS PREFERENCES & RESTRICTIONS
1998 QUEUING MODEL AT SHOPS
POWER CYCLES
PICK-UP AND SET-OFF
1999 CONNECT TO MPS
LEADER LOCOMOTIVE
TRAIN PRIORITIES
MANY UPGRADES (DATA, REPORTS, GUI)
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
HISTORY OF LDM DEVELOPMENT & IMPLEMENTATION
- cont’d
- TACTICAL MODEL (1997 - )
1999 TRIAL IMPLEMENTATION
Consist break-up major issue
2000 PARTIAL SOLUTION IS SENT TO MPS
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
DATA ISSUES
DATA MODEL
==============
èQUALITY DEVELOPMENT TIME & COST
MODEL DISCIPLINE IN
=============
èREQUIREMENTS DATA CREATION PROCESS
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
DATA ISSUES
EXAMPLES :--
- Train Plan (multiple sources - SRS/TOPC/MPS)
- Forecast of Trains for model time horizon
- Forecast of train tonnages (did not exist)
- Work Trains
- Locomotive mechanical status (free form vs code)
- Disconnected Network
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
PROCESS ISSUES
Changing business rules
è shifting model requirementse.g. local vs mainline fleets
leader unit on U.S. trains
Hard vs Soft Rules
e.g. trip servicing deadline
HP to Ton requirement
Lack of Consensus on how model will be used
èArchitecture is evolving
(function of confidence level)
IMPLEMENTING A LOCOMOTIVE OPTIMIZATION MODEL
CANADIAN NATIONAL RAILWAYS
LESSONS LEARNED ?
OR
AWARENESS OF RISK FACTORS
- Detailed analysis of requirements no guarantee that model specs will be properly defined
why?
Long development time means
Players will change
Business rules will change
Difficult to see all "problems parameters" in advance
- Need agreed upon process to deal with data problems
should it be fixed? rules to fix it, who should fix it
- Confidence in model depends on ability to understand "Why"
- Important to document model specs
- Allow generous time for dealing with data issues