CAREER: Robust Control of Large-Scale Constrained Nonlinear Systems

大多数化学过程是非线性的。在过去，基于线性模型的过程控制技术已经足够，主要是由于相对宽松的环境法规和工业的能力（例如，通过过程过度设计）来设计在线性假设有效的地区运行的过程。然而，严峻的全球经济竞争和严格的环境法规的结合正在将过程操作窗口推向线性假设失效和必须满足操作约束（例如，设备，环境，安全等）的地区。此外，今天的过程本质上是大规模的，因为它们正在变得更加一体化，以提高资本生产率。因此，这些趋势推动了过程工业对改进控制技术的需求，这些技术可以有效地处理受操作约束的大规模非线性过程。计算硬件和软件的快速发展为开发先进的控制技术提供了机会，这些技术可以实时优化过程的盈利能力。计划研究的目标是开发一种新的非线性模型预测控制（NLMPC）技术，该技术解决了实际问题（例如，低在线计算需求），并且在理论上是合理的（例如，鲁棒稳定性和性能保证），用于控制大规模约束非线性系统。本研究涉及理论、仿真研究和工业应用的结合，具体包括以下项目：(1)开发一种计算效率高的MPC算法，该算法可以在具有数十个输入和输出的非线性系统上实现。(2)(2)证明在既有模型不确定性又有约束的情况下闭环系统是渐近稳定的(3)(3)推导出有约束非线性系统鲁棒镇定的充分必要条件。(4)建立大型化工过程对苯二腈（TPN）装置第一性原理模型，并进行模拟研究。(5)在醋酸甲酯反应精馏塔上实现该算法，已知线性控制器在某些操作区域会失效。(6)根据理论成果开发教育软件，通过CACHE进行传播。计划进行几项教育创新：(1)开发互动学习模块，使教师能够更多地关注“基于发现”的教学方法，而不是传统的“我说你听”的教学方法。这些模块将把学生从被动的听众转变为积极的参与者。(2)开发开放式问题，这些问题不仅需要工程知识，还需要决策过程的其他方面。(3)开展项目，让入选的本科生更多地参与前沿工程研究。(4)让PI的学生完成工业作业。

Abstract - Zheng - 9984104Most chemical process are nonlinear. In the past, process control techniques based on linear models had been adequate largely due to relatively loose environmental regulations and industry's ability (e.g., via process over-design) to design processes to operate in regions where assumptions of linearity are valid. However, a combination of stiff global economic competition and strict environmental regulations is pushing process operational windows into regions where assumptions of linearity break down and where operating constraints (e.g., equipment, environmental, safety, etc.) must be satisfied. Furthermore, today's processes are large-scale in nature as they are becoming more integrated to improve capital productivity. Consequently, these trends have driven the process industry's need for improved control technologies, which can deal effectively with large-scale nonlinear processes subject to operating constraints. The rapid development in computational hardware and software has provided the opportunity for developing advanced control techniques, which optimize process profitability in real time.The goal of the planned research is to develop a novel nonlinear model predictive control (NLMPC) technology which addresses practical concerns (e.g., low on-line computational demand) and is theoretically sound (e.g., robust stability and performance guarantees) for control of large-scale constrained non-linear systems. The research involves a combination of theory, simulation studies, and industrial applications and, specifically, consists of the following projects: (1) Develop a computationally efficient MPC algorithm which can be implemented on nonlinear systems with tens of inputs and outputs.(2) (2) Prove that the closed-loop system is asymptotically stable in the presence of both model uncertainty and constraints(3) (3) Derive necessary and sufficient conditions for the robust stabilization of constrained nonlinear systems.(4) Develop a first-principles based model for a terephthalonitrile (TPN) plant, a large-scale chemical process, and carry out simulation studies on it.(5) Implement the algorithm on a methyl acetate reactive distillation column where it is known that linear controllers would fail in some operating regions.(6) Develop educational software based on the theoretical results and disseminate via CACHE.Several educational innovations are planned:(1) Develop interactive-learning modules, which allow the instructor to focus more on the "discovery-based" teaching method than the traditional "I-talk-you-listen" teaching method. The modules would transform the students from passive listeners to active participants.(2) Develop open-ended problems which require not only engineering know-how, but also other aspects of the decision-making process.(3) Develop projects to get selected undergraduate students more involved with frontier engineering research.(4) Have the PI's students take industrial assignments.