Clustering methods for control-relevant decomposition of complex process networks

1605549 PI：Daoutidis标题：用于复杂过程网络控制相关分解的聚类法复杂过程网络由许多反应、分离和热交换单元相互连接组成，在现代化工和能源工厂中非常常见。有效地控制这类网络是一个具有挑战性的问题，需要发展分布式控制策略。为了制定这样的战略，必须确定能够在整个流程网络内有效控制和协调的组成的子网络。然而，目前还缺乏一个严格的面向控制的网络分解框架，该框架可以自动化并应用于大规模网络。本研究的主要目的是：i)发展复杂过程网络控制相关分解的通用方法，以及(Ii)将这些方法应用于过程和能源行业的典型系统。该项目将采用网络理论中广泛使用的层次聚类方法，作为分析复杂过程网络的互连性和设计其控制结构的有力框架。在这一框架内，将开发与控制相关的综合过程网络分解的基本方法。这些方法将能够对完全解析的输入/输出集群层次结构进行系统分类，范围从单个集群到输入/输出对的单独集合。他们还将根据紧凑性和紧密性的适当测量来确定这种集群的最佳模块化。因此，开发的网络分解方法将促进复杂过程网络从完全分散的控制范例向更有效的分布式范例的转变。这种高效的全厂控制策略对经济可行性以及化工和能源行业的能源和环境可持续性至关重要。拟议的研究将为研究生在基础研究、交叉数学和控制理论方面的有效培训提供一个环境，并具有及时和重要的应用组成部分。研究成果将通过出版物和演示文稿广泛传播，而将开发的开放源码软件将进一步加强研究和教育的基础设施。

1605549 PI: DaoutidisTitle: Clustering methods for control-relevant decomposition of complex process networksComplex process networks, consisting of interconnections of numerous reaction, separation and heat exchange units, are very common in modern chemical and energy plants. Effectively controlling such networks is a challenging problem that requires the development of distributed control strategies. In order to develop such strategies, constituent sub-networks must be identified that can be effectively controlled and coordinated within the overall process network. However, a rigorous control-oriented network decomposition framework that can be automated and applied to large-scale networks is currently lacking. The main goals of the proposed research are: i) to develop generalized methods for control-relevant decomposition of complex process networks, and (ii) to apply these methods to representative systems from the process and energy industries.The proposed project will employ hierarchical clustering methods that have been used extensively in network theory as a powerful framework for analyzing the inter-connectivity of complex process networks, and designing control structures for them. Within this framework, fundamental methods for control-relevant decomposition of integrated process networks will be developed. These methods will enable a systematic classification of fully-resolved input/output cluster hierarchies, ranging from a single cluster to individual collections of input/output pairs. They will also determine the optimal modularity of such clusters on the basis of appropriate measures of compactness and closeness. The developed network decomposition methods will therefore facilitate the transition from a fully decentralized control paradigm towards a more effective distributed paradigm for complex process networks. Such efficient plant-wide control strategies are critical to the economic viability, as well as the energy and environmental sustainability of chemical and energy industries. The proposed research will provide a setting for the effective training of graduate students in fundamental research, cutting across mathematics and control theory, with a timely and important application component. The research results will be broadly disseminated through publications and presentations, whereas the open-source software that will be developed will further enhance the infrastructure for research and education.