CAREER: Design of in-line controllers for continuously operating networks with structural uncertainty

本项目的重点是通过非侵入性观察来设计具有未知结构的网络系统的控制机制。通过非侵入性观察，这意味着所测量的不是系统对主动注入输入的反应，而是系统在标准条件下运行时的行为，并受到潜在的不可观察的强迫信号的影响。仅基于非侵入性观察设计控制器的能力对于实现关键或不间断功能的任何大规模网络（即电网，物流系统）或在不实际或过于昂贵的情况下将已知探测信号注入系统（即基因网络，金融网络）至关重要。其他相关的应用在医学上（例如，重复药物测试，计算机辅助麻醉）。事实上，在这些情况下，出于明显的安全和健康考虑，如果可以从非侵入性观察中推断出相当有用的信息，则不希望主动测试患者对不同药物剂量或治疗的反应。由于非侵入性观察并不总是提供有关网络配置的完整信息，因此该项目还将考虑如何定义适当的控制机制，这些机制对于连接结构中的不确定性具有鲁棒性。在控制设计的标准技术中通常不考虑这些不确定性，需要开发特定的方法。结合适应网络配置变化的能力，这些控制技术将为实现自愈系统提供坚实的基础。这个项目将把不同的领域，包括统计学、计算机科学和控制理论，用一个统一的框架连接起来。将开设新的课程，以促进所有这些研究人员社区之间的交流，以多学科的方式推进各自的领域。

This project focuses on designing control mechanisms for a networked system with unknown structure by making use only of non-invasive observations. By non-invasive observations, it is meant that what is being measured is not the system reaction to actively injected inputs, but rather the system behavior when it is operating under standard conditions and subject to potentially unobservable forcing signals. The capability of designing controllers based only on non-invasive observations is of paramount importance for any large scale network fulfilling critical or uninterruptible functions (i.e., a power grid, a logistic system) or in situations where it is impractical or too expensive to inject known probing signals into the system (i.e., a gene network, a financial network). Other relevant applications are in medicine (i.e., repeated drug testing, computer- assisted anesthesia). Indeed, in these cases, for obvious safety and health concerns, it is not desirable to actively test the response of a patient to a different drug dosage or treatment, if comparably useful information could be inferred from non-invasive observations.Since non-invasive observations do not always provide full information about the network's configuration, the project will also consider how to define adequate control mechanisms that are robust with respect to uncertainties in the connectivity structure. These kinds of uncertainties are not typically considered in standard techniques for control design and the development of specific methodologies is required. Combined with the capability of adapting to changes in the network's configuration, these control techniques will provide a solid foundation for the realization of a self-healing system. This project will bridge together different areas, including statistics, computer science, and control theory with a single unifying framework. New courses will be created to facilitate communication among all these communities of researchers, advancing separate fields in a multidisciplinary way.