377. Stochastic Multicommodity Network Design: A Synergistic Approach Using Machine Learning and Optimization
Invited abstract in session TC-17: Service Network Design: challenges and opportunities, stream Combinatorial Optimization.
Tuesday, 12:30-14:00Room: Esther Simpson 2.08
Authors (first author is the speaker)
| 1. | Fatemeh Sarayloo
|
| 2. | Mahya Hemmati
|
| Analytics, Operatons, and Information Technologies, Université du Québec à Montréal | |
| 3. | Teodor Crainic
|
| CIRRELT | |
| 4. | Walter Rei
|
| CIRRELT and UQAM |
Abstract
This paper introduces a novel Machine Learning (ML)-guided matheuritics for the stochastic multicommodity capacitated network design problem (SMCND), with broad applications in service network design, transportation, and beyond. Our approach synergistically integrates machine learning and optimization to address both demand uncertainty and computational complexity. We propose an iterative framework that employs Artificial Recourse Problems (ARPs) defined across various scenario combinations to generate a rich dataset, including features related to promising arcs, paths, and costs, which are used to train ML models capable of predicting key solution structures. We explore various algorithmic strategies that facilitate an iterative exchange of information between ML and optimization, guiding the search process toward high-quality solutions. Extensive experiments highlight the gains over traditional methods, especially for large-scale instances.
Keywords
- Combinatorial Optimization
- Machine Learning
- Network Design
Status: accepted
Back to the list of papers