Moving Horizon Estimation and Nonlinear, Large-Scale Model Predictive Control of Chemical Processes

摘要：詹姆斯B。罗林斯研究所：为了保持经济竞争力，最大限度地提高产品质量和运行更环保的过程，化学过程工业需要投资于过程监测和控制技术。 工业过程操作正变得越来越集成和复杂，针对约束的操作是常见的，并且动态过程行为需要更复杂的在线分析和决策，以便优化过程性能。 本研究的目标是解决三个突出的挑战，在进一步发展先进的化学控制方法：1。基于动态模型和在线测量的状态估计用于进一步改进过程操作。 滚动时域估计（MHE）方法解决了这一需求，因为它允许非线性模型和估计器约束作为问题制定的一部分。 PI计划研究一系列问题和应用问题，以理解和扩展MHE理论当前的适用性。使用模型预测控制（MPC）对集成的大型工厂进行在线优化。 PI计划利用MPC预测器中预测范围提供的固有通信能力。开发一套闭环控制的非线性MPC测试问题。 这些问题将为研究界提供一个方便的标准来测试新的非线性MPC理论、方法和算法。这项研究的预期影响包括：提高对在线使用模型进行监测、推理和决策的基本问题的理解;通过开发和发布高级源代码，将新的研究成果迅速广泛地传播给教育工作者。

ABSTRACTPI: James B. Rawlings Institution: University of WisconsinProposal Number: 0105360To maintain economic competitiveness, maximize product quality and operate more environmentally benign processes, the chemical process industries need to invest in process monitoring and control technology. Industrial process operations are becoming increasingly integrated and complex, operation against constraints is common and the dynamic process behavior requires more sophisticated online analysis and decision making in order to optimize the process performance. The goal of this research is to address three outstanding challenges in the further development of advanced chemical control methodology:1. State estimation based on dynamic models and online measurements is used to further improve process operations. The moving horizon estimation (MHE) approach addresses this need as it allows both nonlinear models and estimator constraints as part of the problem formulation. The PI plans to study a series of issues and application problems in order to understand and expand the current applicability of MHE theory.2. Use model predictive control (MPC) for online optimization of an integrated, large-scale plant. The PI plans to exploit the inherent communication capability provided by the forecast horizon in the MPC controllers.3. Develop a suite of nonlinear MPC test problems for closed-loop control. These problems should provide a convenient standard for the research community to test new nonlinear MPC theories, approaches and algorithms.Expected impacts for this research include: improved understanding of the fundamental issues surrounding the online use of models for monitoring, inference and decision making; direct impact on industrial plant operations and control; and rapid and wide dissemination of new research results to educators by the development and release of high-level source code.