CPS: Small: Recovery Algorithms for Dynamic Infrastructure Networks

大多数关键基础设施已经演变成由大量相互作用的要素组成的复杂系统。这些相互作用导致中断的传播，例如延迟，从系统的一个部分到另一个部分，甚至从一个基础设施到另一个基础设施。分析和控制现实世界基础设施的有效工具需要考虑潜在的动态。本研究的关键见解是，通过学习基础设施网络的数据驱动模型，并使用这些模型来确定动态感知恢复算法，我们可以极大地提高关键基础设施网络的弹性。我们建议通过以下方式来应对这些挑战：1.从数据中学习和验证真实系统的可伸缩表示。通过考虑连续状态，并通过将连通性的时变特性建模为网络拓扑之间的切换，我们提出了一类切换线性系统模型。将开发多层网络模型，以考虑航空公司网络和多式联运系统。表征作为整体的系统的弹性，以及关于单个节点的弹性(例如，对网络延迟的敏感度)。评估复原力的指标将包括稳态行为和暂态行为。使用已识别的模型来设计能够实现中断恢复的最佳控制算法，同时考虑网络动态、操作环境的不确定性以及在不同级别、不同时间做出恢复服务的决策的成本。研究结果将使用操作数据进行验证，从而产生一套用于系统诊断、分析和恢复的工具。改善和维护关键基础设施是美国国家工程院确定的重大挑战之一。这项拟议的研究将发展基于网络科学、机器学习以及系统和控制理论的技术，以便有效地设计和运营基础设施。为这些基础设施制定共同框架和抽象将有助于研究它们之间的相互依存关系。随着智能基础设施的快速发展，拟议的研究将造福社会，也有助于吸引和培养下一代工程专业人员。

Most critical infrastructures have evolved into complex systems comprising large numbers of interacting elements. These interactions result in the spread of disruptions, such as delays, from one part of the system to another, and even from one infrastructure to another. Effective tools for the analysis and control of real-world infrastructures need to account for the underlying dynamics.The key insight in this research is that by learning data-driven models of infrastructure networks, and using these models to determine dynamics-aware recovery algorithms, we can greatly improve the resilience of critical infrastructure networks. We propose to address these challenges by:1. Learning and validating scalable representations of real systems from data. By considering continuous states, and by modeling the time-varying nature of connectivity as switching between network topologies, we propose to obtain a class of switched linear system models. Multilayer network models will be developed to account for airline networks, and multimodal systems.2. Characterizing resilience, both for the system as a whole, and in terms of individual nodes (e.g., susceptibility to network delays). The metrics to evaluate resilience will encompass both steady-state and transient behavior.3. Using the identified models to design optimal control algorithms that can enable recovery from disruptions, taking into account network dynamics, the uncertainty in operating environments, and the costs of decisions to restore service at various levels, at various times.The results of the research will be validated using operational data, thereby yielding a set of tools for system diagnostics, analysis, and recovery. Improving and maintaining critical infrastructures are among the grand challenges identified by the National Academy of Engineering. The proposed research will develop techniques grounded in network science, machine learning, and systems and control theory in order to effectively design and operate infrastructures. The development of common frameworks and abstractions for these infrastructures will enable the study of their interdependencies. With the rapid growth of intelligent infrastructures, the proposed research will benefit society, and also help attract and train the next generation of engineering professionals.