Statistical and machine learning models for critical infrastructure resilience

This thesis presents a data-driven approach to improving predictions of critical infrastructure behaviors. In our first approach, we explore novel data sources and time series modeling techniques to model disaster impacts on power systems through the case study of Hurricane Sandy as it impacted the state of New York. We find a correlation between Twitter data and load forecast errors, suggesting that Twitter data may provide value towards predicting impacts of disasters on infrastructure systems. Based on these findings, we then develop time series forecasting methods to predict the NYISO power system behaviors at the zonal level, utilizing Twitter and load forecast data as model inputs. In our second approach, we develop a novel, graph-based formulation of the British rail network to model the nonlinear cascading delays on the rail network. Using this formulation, we then develop machine learning approaches to predict delays in the rail network. Through experiments on real-world rail data, we find that the selected architecture provides more accurate predictions than other models due to its ability to capture both spatial and temporal dimensions of the data.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Aerospace Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2020

Author dc:creator

• Heglund, Jacob Scott White

Contributors dc:contributor

• Tran, Huy T

• Critical Infrastructure Resilience
• Machine Learning
• Statistical Modeling
• Graph Neural Networks

Statement dc:rights

• Copyright 2020 Jacob Heglund

Language dc:language

en

Handle dc:identifier

http://hdl.handle.net/2142/108533

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/108533