Large-Scale Discrete-Time Scheduling Optimization: Industrial-Size Applications

Abstract

Optimization of large-scale discrete-time scheduling problems is challenging due to the combinatorial complexity of binary or discrete decisions to be made. When including networks of unit-operations and inventory-tanks to fulfill both the logistics and quality balances as found in complex-scope process industries, the decomposition of mixed-integer nonlinear programming (MINLP) regarding its quantity-logic-quality phenomena (QLQP) paradigm into mixed-integer linear programming (MILP) and nonlinear programming (NLP) has been commonly and naturally used to find solutions of industrial-sized problems. Other approaches can be incorporated into an optimization-based decision-making framework to provide proper capabilities for handling complex large-scale applications. This includes strategies related to reduction of model, time, and scope that can be based on machine learning approaches and heuristic algorithms. Such a decision-making framework is useful not only to allow solving industrial-scale problems, but also to achieve enhanced applications. There are open challenges to automatically solve complex large-scale discrete-time problems in acceptable computing time. In this context, this paper employs a decision-making framework based on modeling and optimization capabilities to handle large-scale scheduling problems. The examples are built using the unit-operation-port-state superstructure (UOPSS) constructs and the semantics of the QLQP concepts in a discrete-time formulation. The proposed framework is shown to effectively use decomposition and heuristic strategies for solving industrial-sized scheduling formulations.