Data-driven robust MILP model for scheduling of multipurpose batch processes under uncertainty

Abstract

Robust optimization has been widely used in the scheduling of multipurpose batch processes under uncertainty. However, traditional robust scheduling methods make simple assumptions about uncertainty, such as independence and symmetry. This paper proposes a novel scheduling approach of batch processes based on a data-driven robust mixed-integer linear programming (MILP) model. The Dirichlet process mixture model is adopted to construct an uncertainty set via variational inference from the historical data of uncertainty parameters. A data-driven robust counterpart of a general MILP is then derived as a conic quadratic mixed-integer programming based on this uncertain set. A specific data-driven robust MILP model is further developed to address multipurpose batch process scheduling problem under various uncertainties. An industrial case study is presented to demonstrate the effectiveness of the proposed method.