Heuristic Data-Driven Feasibility on Integrated Planning and Scheduling

Abstract

We study the merging of data-driven approaches and mathematical programming formulations to solve an integrated planning and scheduling problem where jobs can be split in two separate tasks, one of them allowed to exceed its deadline at a price. Our study is driven by the increasing structural complexity of industrial scheduling problems that in some cases become too hard to be modeled as mathematical programs even by domain experts. We experiment on how to ensure the feasibility at a scheduling level by training a data-driven model, subsequently encoding it with a mathematical programming formulation, to be finally embedded in a planning model. Our experiments prove that our framework provides an effective heuristic approach, competing to exact formulations in terms of both accuracy and quality of the solutions, and it could be extended to those kind of problems where it is too hard to model the schedule feasibility.