Reproduce “Uncovering expert objectives in production planning via inverse optimization: An industrial case study” on OpenResearch

About this paper

Shivi Dixit, Rishabh Gupta, Adam Kelloway, John Wassick, Qi Zhang

Inverse optimization is used to recover the implicit priorities behind industrial production plans while retaining the plant’s detailed MILP constraints. On 70 rolling-horizon plans for an eight-product facility, models incorporating inventory targets, cycle-length consistency, and production gaps predict planner decisions more accurately than an inventory-cost baseline and reveal that avoiding shortages matters more than minimizing excess inventory.

Key insights

• The learned objective emphasizes maintaining inventory above desired lower bounds, indicating that shortage and customer-service risks outweigh the penalty associated with excess inventory.
• Cycle-length regularity is an important planning consideration, while the inferred inventory holding-cost weight is zero in the richer model—consistent with expert feedback that planners use target ranges rather than explicit holding-cost calculations.
• Time-dependent weights reflect rolling-horizon behavior: near-term decisions prioritize inventory protection, whereas later decisions place more emphasis on stable cycle patterns; product-specific weights further improve fit but may also capture demand or forecast differences.

Summary by alphaXiv.