Tractable Data-driven Solutions to Hierarchical Planning-scheduling-control

Using numerical optimization for the hierarchical integration of decision-making units is crucial to provide feasibility and optimality of all levels. However, realistically modelling hierarchical decision-making calls for multilevel formulations, which are numerically intractable and mathematically difficult. In this work, we show how to leverage two data-driven techniques – derivative-free optimization and optimality surrogates – to decrease the computational burden of multilevel problems. We reformulate a tri-level planning-scheduling-control problem into a single-level black-box problem wherein each evaluation calls a scheduling instance with embedded optimal control surrogates. We show that solving this integrated problem instead of the single-level instance leads to changes in the optimal production planning and scheduling sequence, and discuss trade-offs associated with both techniques.