Development of a novel predictive controller synthesis method for complex reaction systems

PI：Armaou，Antonios研究所：宾夕法尼亚州立大学提案编号：1264902标题：为复杂反应系统开发一种新的预测控制器合成方法智能优点：利润率的持续压力和日益严格的环境限制导致需要开发先进的控制结构，迫使优化过程行为，同时满足严格的过程和产品质量限制。在过去的二十年里，模型预测控制(MPC)已成为化工行业广泛使用的有力工具。MPC的控制力是通过在线反复求解一个有限时间开环优化问题来计算的，由于控制力是在过程演化过程中计算出来的，因此在强迫系统遵循一定的尊重过程约束的最优路径的过程中，MPC具有抑制外部干扰和容忍模型误差的能力。这项工作旨在将MPC设计的适用性扩展到复杂过程，并解决它们的计算需求，这些问题到目前为止阻碍了MPC在快速演化和不稳定过程中的实现。智力的目标是开发一个通用的和系统的MPC合成方法，专门为化学和能源领域的过程量身定做。这项工作将解决与a)最优控制问题的动态性质有关的基本计算问题，以及b)由于交通现象和反应的相互作用而引起的空间变化。个别项目的目标包括：o开发一种计算效率高的算法，以导出描述动态过程行为的非线性低阶、近似的代数模型。o错误的表征和用户定义的误差范围的实施。o通过将基本的动态优化问题重新表述为符合标准搜索算法的代数问题，构建实际可实施的MPC设计。o MPC设计的计算加速。模型的非线性和不确定性对MPC结果的影响。o将研究结果整合到研究生课程和软件工具的传播中；本科生参与研究和修改研究生课程的选定部分。更广泛的影响：广泛的复杂工业过程可以从这项研究的结果中受益。例如，用于生产不饱和聚酯的间歇反应器和反应蒸馏塔，以及微电子制造工艺(例如，气相外延、化学气相沉积、蚀刻和电沉积)。后一种工艺被广泛用于生产有机和无机光伏系统。PI旨在将研究成果应用到实际工业过程中，重点关注经济运行和对关键过程和产品特性的严格控制。预测控制器将用于与NTUA和PSU实验者合作，设计更高效的实验，发现关键过程参数，并识别最优的依赖时间的操作条件。作为回报，开发的方法将在真实的实验反应堆中进行评估；相关的弱点将被识别和解决。为了将研究的结果和见解转移到工业部门，PI还将积极寻求与业界的合作，并将开发和传播具有透明的用户-机器交互界面的软件工具。此外，PI计划开展一系列活动，将研究与教育相结合，包括将研究成果纳入优化和控制课程，本科生通过荣誉计划参与研究，以及开发教育工具。最后，PI将利用目前可用的场所向宾夕法尼亚州立大学的其他研究小组、其他教育机构和行业有效地传播该软件。

PI: Armaou, AntoniosInstitutions: Pennsylvania State UniversityProposal Number: 1264902Title: Development of a novel predictive controller synthesis method for complex reaction systemsIntellectual Merit: Continuous pressure on profit margins and ever stringent environmental limits have ledto a need to develop advanced control structures that force optimal process behavior and simultaneouslysatisfy strict process and product quality constraints. Over the past twenty years model predictivecontrol (MPC) has become a powerful tool that is extensively used by the chemical industry. The controlaction in MPC is calculated by repeatedly solving online a finite-horizon open-loop optimization problem.As the control action is computed during process evolution, MPC has the capability to suppress the externaldisturbances and tolerate model inaccuracies during the course of forcing the system to follow a certain optimalpath that respects the process constraints. Issues that significantly limit the practical implementationof MPCs is their computational requirements, the need for development of specialized search algorithms andperformance degradation due to model uncertainty.Motivated by the above, this work aims to extend the applicability of MPC designs to complexprocesses and address their computational requirements which have so far prevented their implementation tofast evolving and unstable processes. The intellectual objective is to develop ageneral and systematic MPC synthesis methodology that is specifically tailored for processes in the chemicaland energy fields. The work will resolve fundamental computational issues associated with a)the dynamic nature of optimal control problems, and b) spatial variations due to the interplay of transportphenomena and reaction. Individual project aims include:o Development of a computationally efficient algorithm to derive nonlinear low-order, approximatealgebraic models that describe the dynamic process behavior.o Characterization of error and enforcement of user-defined error bounds.o Construction of practically implementable MPC designs via reformulation of the underlying dynamicoptimization problem as an algebraic one that is amenable to standard search algorithms.o Computational acceleration of the MPC designs. Characterization of model nonlinearity and uncertaintyeffects on MPC results.o Integration of the research results into the graduate curriculum and dissemination of softwaretools; involvement of undergraduate students into selected parts of the research and revision ofundergraduate curriculum.Broader Impact: A wide range of complex industrial processes could benefit from the results of this research. Examples include both batch reactors and reactive distillation columns for the production of unsaturated polyesters, and microelectronics manufacturing processes (e.g. vapor phase epitaxy, chemical vapor deposition, etching & electrodeposition). The latter processes are extensively used forthe production of both organic and inorganic photovoltaic systems. The PI aims to implement the research results in real-life industrial processes focusing on economic operation and tight control of key process and product characteristics.The predictive controllers will be used in collaborative efforts with NTUA and PSU experimentaliststo design more efficient experiments, discover crucial process parameters and identify optimaltime dependent operating conditions. In return, the developed methodologies will be evaluated in real-lifeexperimental reactors; relevant weaknesses will be identified and addressed.To transfer the results and insight of the research to the industrial sector, the PI will also activelyseek collaborations with industry and will develop and disseminate software tools with a transparentuser-machine interaction interface. In addition, The PI plans a number of activities to integrate theresearch with education including incorporation of research results in optimization and control courses,undergraduate student participation in research through the honors program, and the development of educationaltools. Finally, the PI will employ current available venues to efficiently disseminate the software toother research groups within Penn State, other educational institutions and industries.