Parameterizations of data-driven nonlinear dynamic process models for fast scheduling calculations

Abstract

Global competition and increasingly complex product slates and supply chains motivate a continuous drive towards enterprise-wide optimization and integrated decision-making in the chemical process industries. Integration of production scheduling and process control poses particular challenges: the resulting optimization problems tend to be high-dimensional and nonlinear, calling for development of new computational methods. In this work, we propose a novel modeling framework for integrated scheduling and control. We build on existing methods which use data-driven Hammerstein–Wiener models to represent the dynamics of (scheduling-relevant) process variables. This model structure is leveraged to reduce the size of the scheduling optimization problem, by identifying parsimonious parametric representations of the underlying dynamics. The advantages of the approach are demonstrated on two case studies, in which the computational effort is shown to be significantly reduced compared to existing methods, while still capturing the relevant process dynamics.