CAREER: Decentralization and Parsimony for Implementable Control of Massively Interconnected Systems

目的：复杂互联系统的出现产生了设计和分析控制器的需求，这些控制器只能从网络的一小部分观察信息，但最终可能影响网络的大部分。这包括智能建筑管理、多车辆系统和车队、灌溉网络、大型阵列望远镜和配电电网，这是网络物理系统许多问题中的关键挑战。传统的控制分析假设一个集中的决策者可以访问所有可用的测量，并确定所有可能的激励手段的使用。大多数设计和分析方法对这一假设极其脆弱，当这种集中化不可能或不希望时就会崩溃，从而导致分散控制领域。智力优势：目前，在分散控制方面，著名的理论进步与用于实现甚至计算的概念之间存在着巨大的脱节。无论是最近的进展还是更经典的结果，都是如此。该项目探讨了这种脱节的主要原因，以及系统实施分散控制理论的其他迫在眉睫的障碍，特别是那些在应用于大规模系统时将变得无效的障碍。它进行理论调查，旨在以一种可以消除这些障碍的方式推动该领域的发展，并带来更深远的好处，再加上旨在利用过去和未来在使技术能够用于上述敏感应用方面的进展的计算和算法调查。更广泛的影响：该项目将产生节俭恢复理论和方法的新综合，该理论和方法最近经历了戏剧性的发展，在分散控制方面既有经典的结果，也有现代的进步。它将进一步扩大优化理论的优雅和有用方面的适用性，使其适用于对项目主要范围至关重要的各类问题。在优化和估计方面追求的根本进步有可能更广泛地使用，并影响许多其他领域。该项目还寻求通过与工业界和实验者的合作，以及通过创建供非专家广泛使用的软件工具，在其主要领域之外产生广泛的影响。

Objective: The advent of complex interconnected systems has created a need to design and analyze controllers that can observe information from only a small portion of a network but may ultimately effect a large portion of the network. This includes smart building management, multi-vehicle systems and convoys, irrigation networks, large array telescopes, and the power distribution grid, and is a key challenge in many problems with cyber-physical systems. Conventional controls analysis assumes that one centralized decision-maker can access all available measurements, and determine the usage of all possible means of actuation. Most methods of design and analysis are extremely fragile to this assumption, and break down when such centralization is not possible or is not desired, leading to the field of decentralized control. Intellectual Merit: There is currently an enormous disconnect in decentralized control between the celebrated theoretical advances and the concepts that are used for implementation, or even for computation. This is true of both recent advances and more classical results. This project pursues the key reasons for this disconnect, along with other impending barriers to the systematic implementation of decentralized control theory, particularly those which will become disabling when applied to massive systems. It undertakes theoretical investigations targeted to advance the field in a manner from which those barriers can be eliminated, along with much farther-reaching benefits, further coupled with computational and algorithmic investigations designed to parlay past and future advances into enabling technologies for sensitive applications including those listed above. Broader Impacts: This project will produce a novel synthesis of the theory and methods of parsimonious recovery, which has undergone dramatic recent developments, with both the classical results and modern advances in decentralized control. It will further broaden the applicability of elegant and useful aspects of optimization theory to classes of problems that are paramount for the main scope of the project. The fundamental advances pursued in optimization and estimation have the potential to be of use much more broadly and to impact many other fields. This project further seeks to make broad impacts outside of its primary domain through collaborations with industry and with experimentalists, and through the creation of software tools for widespread use by non-experts.