Best Achievable Control System Performance

摘要：近年来，过程工业缺乏重大资本投资，使实施控制项目成为提高工厂盈利能力的一种模式。这些经济效益通常与先进控制提供的改进的动态性能有关。在法规控制上下文中，希望质量变量保持在控制范围内，不允许超过规格限制。如果一种控制策略可以达到比另一种控制策略更严格的控制极限，则可以改变该质量变量的设定值，从而提高过程经济性。动态性能不是选择控制系统的唯一标准。不确定性、复杂性和可靠性也是关键问题。化学过程模型往往是经验的，忽略了复杂的物理化学现象或缺乏过程参数的信息。因此，高度的不确定性与这些模型有关。化学过程也是大规模的相互关联的系统，其中大量的变量被测量和/或操纵。这些相互联系的单位作业的地理位置需要安装一个复杂的通信网络，以便将信息从测量处传送到一个中心位置，然后再传送到操作处。安装和维护成本以及这种网络的复杂性往往阻碍了这种集中控制系统的实施。相反，控制系统的简单性是通过以下方式实现的：1)使用分散控制系统，其中测量和操作被分类到子集中，并且仅在相应子集之间交换信息；20使用简单的低阶控制策略，这是人类操作员所熟悉的。可靠性是控制系统设计的另一个标准。控制系统元件（如传感器/执行器）的故障常常导致控制回路的手动操作。理想的控制系统是这样的，它可以很容易地重新配置，以维持工厂的运行，尽管有退化的动态性能特征。本研究项目的目标是开发一种方法来解决以下问题：在存在外部干扰的情况下，分散和/或低阶同时稳定和/或可靠的反馈控制系统可以提供的最佳动态性能是什么？这些控制问题的表述导致了二次约束、无限维优化问题。研究这些问题的全局解决的计算技术是将要讨论的主要问题之一。将寻求分支定界、分解、区间算术技术及其组合和计算实现。

Abtract - Manousiouthakis - 9312489 In recent years, the lack of major capital investments in the process industries has made the implementation of control projects one mode to increased plant profitability. These economic benefits are typically associated with the improved dynamic performance offered by advanced controls. In a regulatory control context, a quality variable is desired to stay within control limits and is not allowed to exceed specification limits. If one control strategy can achieve tighter control limits than another, then the setpoint of this quality variable may be changed so as to enhance process economics. Dynamic performance is not one criterion in selecting a control system. Uncertainty, complexity and reliability are also critical issues. Chemical process models are often empirical, neglect complex physicochemical phenomena or lack information on process parameters. Consequently, high levels of uncertainty are associated with these models. Chemical processes are also large scale interconnected systems in which a large number of variables are measured and/or manipulated. The geographical location of these interconnected unit operations would necessitate the installation of a complex communication network for the purpose of transferring information from the measurements to a central location and then to the manipulations. The installation and maintenance costs and the complexity of such a network often prevent the implementation of such a centralized control system. Instead, control system simplicity is attained by: 1) the use of a decentralized control system in which measurements and manipulations are classified into subsets and information is exchanged only between corresponding subsets, and 20 the use of simple low order control strategies with which human operators are familiar. Reliability is another criterion in control system design. Failures of control system elements such as sensors/actuators often lead to the manual operation of control loops . It is desirable that the control system be such that it can be readily reconfigured so as to maintain plant operation albeit with degraded dynamic performance characteristics. The objective of this research project is the development of a methodology that addresses the question: What is the best achievable dynamic performance that a decentralized and/or low order simultaneously stabilizing and/or reliable feedback control system can deliver in the presence of external disturbances. The formulation of these control problems leads to quadratically constrained, infinite dimensional optimization problems. The investigation of computational techniques for the global solution of such problems is one of the main issues that will be addressed. Branch and bound, decomposition, interval arithmetic techniques and combinations thereof and computational implementations will be sought.