Hybrid Column Generation Heuristics with Reinforcement Learning to Solve the Network Planning Problem with Least Path Constraints

Cities are facing significant challenges related to network infrastructure, transportation, logistics, and resource distribution. As an example, we will address the two-level problem called the Minimum Path Constraint Network Planning Problem (FCNDP-SPC), which deals with the routing of multiple products from their origins to their destinations. Each product must be transported by the shortest possible path, considering the length of the arches in the created network. To find better solutions, we integrated column generation with reinforcement learning, using the learning to adaptively determine when to proceed with soft-fixing. The positive results found strengthen the contribution of this study to the advancement of the knowledge of hybrid methods.