Towards a Theory of Closed-Loop Robustness for Real-Time Learning and Control of Multilayered Networks under Cascading Failures
Abstract
Multilayered infrastructure networks (e.g., an inter-dependent system consisting of control, communication and physical layers) are often the targets of WMD attacks because of the catastrophic damage it may cause. For example, some recent studies suggest that even a small-scale attack on the U.S. power grid could cause a nation wide blackout. The state of the art on network robustness of multilayered networks primarily focuses on cascading failures based on the network topology before the attack, aiming to optimize the worst-case performance. However, during and after a WMD attack, the multilayered network often exhibits the following two unique characteristics. First (Imperfect Network Knowledge), a WMD attack will cause a massive loss of network knowledge at the control layer including the network topology, the routing tables, the channel state information, etc. The control layer needs to function properly with incomplete and even faulty network knowledge for controlling both the communication and infrastructure layers. In the damage area, the topology and the inter-layer dependency structure might be only described probabilistically or on a coarse granularity. Second (Rapid Dynamics), a multilayered network in the after-attack scenario exhibits rapid dynamics, whose topology and dependency structure could experience dramatic changes over a short time period. Without the real-time sensing and learning on such after-attack characteristics, the current robustness control strategies often lead to sub-optimal outcome for a multilayered network. In response to these challenges, this project focuses on maximizing the survival network capacity via real-time control and intervention. Specifically, it tackles three complementary research tasks. First (modeling), this project develops new multilayered network models to characterize a multilayered network in the after-attack scenario with rapid dynamics and imperfect network knowledge. Second (tracking and attribution), it develops scalable algorithms to track network robustness and identifies its key attributions which causes a dramatic change of network robustness. Third (real-time control), it develops new optimal strategies to control the closed-loop robustness in real-time. This project aims to contribute to C-WMD science by providing (1) the basic understanding of the fundamental limits of multilayered network robustness with imperfect network knowledge and rapid dynamics; and (2) new instruments to model, track and control the robustness of a multilayered network in real-time with catastrophic damages. The theory and algorithms in this project are designed to answer the following three basic scientific questions. First (Q1), what are the fundamental limits of the network robustness with the imperfect network knowledge and rapid dynamics? Second (Q2), what is the most valuable network information for the control layer to collect for preserving maximal robustness? Third (Q3), how to design robust and scalable learning and control algorithms under noisy and ambiguous network information?
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- May 26, 2016
- Source ID
- HDTRA11610017
Entities
People
- Hanghang Tong
Organizations
- Arizona State University
- Defense Threat Reduction Agency