Title: Data-Driven Methods for Integrated Production Scheduling and Process Control
Speaker: Calvin Tsay
Affiliation: McKetta Dept of Chemical Engineering, University of Texas at Austin
Location: 218 Huxley Building
Time: 11:00 – 12:00
Abstract. Due to the fast-changing market conditions enabled by globalization and modern infrastructures, industrial production scheduling is often performed over relatively short time intervals to maximize profits. For chemical processes, plant dynamics and control become highly relevant at these shortened time intervals, and careful attention is required to ensure computed schedules are feasible when implemented in the physical process. With this motivation, many recent works focused on integrating dynamic information from the process control layer into production scheduling. Unfortunately, the resulting optimization problems are high-dimensional and often intractable because of the broad range of time scales involved.
In this presentation, we describe data-driven techniques for learning low-order dynamic models of the behavior of a process and its controller (i.e., “closed-loop” behavior). Then, we formulate an optimal scheduling problem involving the learned, reduced-order representation of the relevant dynamics. Furthermore, we present a data-mining approach that exploits historical process data to reduce the dimensionality of the aforementioned dynamic models. Throughout the presentation, we will focus on applications to the scheduling of industrial air separation units, which consume immense amounts of electricity, in response to time-varying electricity prices (an operational strategy termed “demand response”). Several case studies will be presented, ranging from model-based analyses to a real-world application on an industrial air separation unit.