Machine Learning-Guided Metaheuristics for Solving Facility Location Problems

Abstract

Facility Location Problems (FLPs) are critical for optimizing logistics, telecommunications, and emergency networks. However, their NP-hard nature limits the scalability of exact resolution methods. While traditional metaheuristics provide effective approximate solutions, they rely on manual expert design. This doctoral thesis addresses this limitation by proposing the integration of Machine Learning (ML) techniques into metaheuristic frameworks to efficiently solve FLPs. The main objective of this thesis is to develop a comprehensive Neural Combinatorial Optimization (NCO) framework based on ML methods to solve optimization problems. By enabling algorithms to autonomously learn resolution strategies directly from network topologies and self-generated data, this work seeks to generate high-quality solutions in reduced computational times. Furthermore, the project aims to evaluate the real-world applicability of these hybrid models, promoting transparent, efficient, and sustainable optimization practices.

Publication
Doctoral Consortium, XXI Conferencia de la Asociación Española para la Inteligencia Artificial (CAEPIA 2026)