(328a) Robustness, Model Predictive Control, and Learning-Based Methods

This presentation provides a perspective review of some of Jay H. Lee's publications on robustness, robust model predictive control, and learning-based methods [1-7].

The review begins by reviewing Jay's work that showed that a well-known and widely used criterion for selecting the best control structure -- the condition number criterion -- could produce poor results when model uncertainties are taken into account [1]. Jay's work inspired many other researchers, and some of the follow-up work is reviewed that provided specific chemical processes that showed that the condition number criterion could select control structures that are completely useless. Jay showed that taking into account the structure of the model uncertainties -- essentially ignored in prior work -- was critical to selecting the best control structure, and he provided computationally efficient screening tools for doing so. His work was a turning point in the study of control structure selection, so much so that the field studied since the 1960s rapidly became mature after the publication of Jay's work.

Next Jay's highly cited paper on min-max formulations of model predictive control with bounded parameters is reviewed [2]. Previous work on min-max formulations was based on the viewpoint of open-loop control, which was the dominant perspective used in model predictive control formulations at the time. His work reconsidered the problem from the perspective of closed-loop control. He demonstrated that min-max MPC based on the open-loop control assumption can give poor closed-loop performance, even when the underlying system is linear time-invariant. He developed two min-max model predictive control formulations, analyzed their closed-loop properties including asymptotic stability of the closed-loop systems, and developed some computationally efficient numerical algorithms. This paper continues to be cited nearly 30 years after its publication.

The presentation closes with an overview of some of Jay's later work on developing methods for incorporating probabilistic uncertainties into optimal control and scheduling, including learning-based methods, and their application to chemical process systems [3-7].

1. J.H. Lee, R.D. Braatz, M. Morari, A. Packard. Screening tools for robust control structure selection. Automatica 31(2), 229-235, 1995.
2. J.H. Lee, Z. Yu. Worst-case formulations of model predictive control for systems with bounded parameters. Automatica 33(5):763-781, 1997.
3. J.H. Lee, K.S. Lee, W.C. Kim. Model-based iterative learning control with a quadratic criterion for time-varying linear systems. Automatica 36 (5), 641-657, 2000.
4. J. Kim, M.J. Realff, J.H. Lee. Optimal design and global sensitivity analysis of biomass supply chain networks for biofuels under uncertainty. Computers & Chemical Engineering 35(9), 1738-1751, 2011.
5. J.H. Lee, K.S. Lee. Iterative learning control applied to batch processes: An overview. Control Engineering Practice 15 (10), 1306-1318, 2007.
6. J.H. Lee, J. Shin, M.J. Realff. Machine learning: Overview of the recent progresses and implications for the process systems field. Computers & Chemical Engineering 114, 111-121, 2018.
7. J. Shin, T.A. Badgwell, K.H. Liu, J.H. Lee. Reinforcement learning--Overview of recent process and implications for process control. Computers & Chemical Engineering 127, 282-294, 2019.