Norfolk Southern plans locomotives for a large rail network, and it wanted a model to help decide how big its fleet should be.
Exact integer-programming models were beyond solver reach and needed assumptions that made the fleet look smaller than it needed to be.
Researchers built an approximate-dynamic-programming model that mixes simulation and optimization with feedback learning, calibrating to historical metrics so it can plan fleet size and mix and test many operating scenarios.
Use simulation plus learning to size a fleet, not a rigid LP
Fleet sizing required a highly detailed model
Integer programming forced simplifications that understated the fleet
Simulation plus learning let the model match real history
The result plans fleet size and mix under many scenarios
When a real problem is too large or too detailed for an exact model, approximate dynamic programming can learn from simulation to size expensive assets accurately.
The model became the way Norfolk Southern studies locomotive fleet size and mix and tests operating changes, a notable use of approximate dynamic programming in freight rail.
FOLLOW THE EVIDENCE
The sources
- Locomotive Planning at Norfolk Southern: An Optimizing Simulator Using Approximate Dynamic Programming Research with NJ / INFORMS Interfaces · researchwithnj.com
- Operations Research at Norfolk Southern Railway Company Springer / Operations Research Forum · link.springer.com