LARGE-SCALE LOGISTICS: COLUMN GENERATION AND STATIC SIFTING IN A ROLLING HORIZON

This paper addresses the tactical planning of a large-scale logistics network, modeled as a Multicommodity Flow Problem with side constraints. Due to the problem's magnitude, reaching billions of variables over the full 30-day horizon, direct methods become infeasible due to memory exhaustion. We propose a hybrid Column Generation approach integrated with a rolling horizon framework, employing a Static Sifting strategy on a pre-computed set of paths. The study investigates insertion strategies, demonstrating that limited multiple column insertion outperforms both conservative insertion (few columns per iteration) and unrestricted generation (all eligible columns). Results in stress scenarios prove that the method reduces dimensionality by approximately 85% and computational time by approximately 20%, ensuring optimality where extensive formulations fail. A heuristic variant (Skip-Pricing) accelerates resolution by up to 34% with marginal solution deterioration.