Back to search

University of Washington

A General Methodology for Inferring Failure Propagation Process from Post-disaster Disruptions Data

Abstract

dc:description.abstract

Cascading failures, where failures propagate from an initially small portion of a system to a much larger portion or even the entire system, are ubiquitous phenomenon in a number of natural, social and technical systems, and have seen instances in the formation of large-scale disasters such as the recent devastations caused by Hurricanes Harvey, Irma and Maria. Understanding and ultimately controlling the dynamics of cascading failures are critical for securing the functionality of those systems, and ensuring the wellbeing of the people and the society. This, however, is a challenging task due to the complexities in the cascading failure dynamics. In this dissertation, a novel methodology to model, infer and reconstruct cascading failure dynamics is proposed, tested and validated. A survival-analysis-based formulation is derived to mathematically describe how external factors and failure propagations give rise to observed failure outcomes, and maximum likelihood estimation is employed to estimate the model parameters. With the estimated parameters, cascading failure dynamics, including the temporal-spatial patterns of failure spreading and node-to-node failure propagation patterns, are reconstructed. This approach is applied to four simulation studies: (a) cascading failures in interdependent power and transportation networks in New York City (NYC) during Hurricane Sandy; (b) a hypothetical influenza epidemic in NYC; (c) a congestion cascade scenario in the Sioux-Falls benchmark network; and (d) cascading power outages in the Wood-Wollenberg 6-bus benchmark system. The inference results returned by the proposed approach and simulation results are compared for each simulation study. All comparisons in the simulation studies return consistent patterns between the inferred cascading failure dynamics and simulated cascading failure dynamics, suggesting the accuracy, robustness and generalizability of the proposed methodology. This dissertation demonstrates strong potential of broad impact both within and beyond the domain of civil engineering. The civil infrastructure systems like transportation and power are constantly under the threat of cascading failures. A methodology for understanding the cascading failure dynamics, especially the node-to-node failure propagation patterns that have been overlooked in existing research, will open a channel to more efficiently enhancing the resilience of infrastructure systems against malfunctioning and assisting emergency response to effectively contain the impact of disruptive events such as natural disasters, terrorist attacks and internal performance fluctuations. More broadly, cascading failures are also observed in a number of other disciplines such as information science, social science, epidemiology, biology and physics. Researchers in those disciplines are faced with similar challenges in learning cascading failure dynamics. The present research constructs a universal methodological framework for understanding and controlling cascading failures, which is applicable to a broad range of systems studied by researchers from various disciplines. It thus will potentially facilitate the inter disciplinary communications that will foster more efficient and high-impact research in each of the individual fields.

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Guan, Xiangyang
Advisor dc:contributor.advisor
  • Chen, Cynthia

Subjects

dc:subject × 8

Rights

dc:rights
Statement dc:rights
  • none
Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
http://hdl.handle.net/1773/43335

Chain of custody

source
Harvested from
University of Washington
Base URL
digital.lib.washington.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Guan, Xiangyang. A General Methodology for Inferring Failure Propagation Process from Post-disaster Disruptions Data. 2019. http://hdl.handle.net/1773/43335