Georgia Tech Coronavirus (Covid-19) Information

PhD Defense by Yanjie Tong

Ph.D. Thesis Defense Announcement

New Approaches for Modeling and Reliability Assessment of Infrastructure Flow Networks

by

Yanjie Tong

 

Advisor(s):

Dr. Iris Tien

 

Committee Members:

Dr. John E. Taylor (GT-CEE), Dr. Samuel Coogan (GT-CEE), Dr. Nagi Z. Gebraeel (GT-ISyE), Dr. Yao Xie (GT-ISyE)

 

Date & Time: April 26th, 2021 | 12-2 pm EST

Location: https://bluejeans.com/911407363

 

Infrastructure flow networks, such as water pipelines, power distribution systems, etc., play an important role in our daily life. These networks are subject to increasing threats and disruptions, both from natural disaster events and more targeted attacks. It is critical to be able to accurately assess the reliability of a network to inform decisions regarding maintenance, retrofitting, etc., to secure the functionality of the network. Infrastructure flow networks are also complex with many components and links operating together. Thus, computational tractability and efficiency is paramount. In this thesis, we propose new approaches for modeling and reliability assessment of infrastructure networks, investigating the reliability in terms of both connectivity and flow capacity. The proposed approaches enable accurate and computationally efficient assessment of infrastructure flow networks.

The first step in the investigation on network reliability considers inputs to a network model with the objective of predicting values characterizing the nodes in a network. We begin with collected data from the field for data preprocessing. We propose a pairwise gated recurrent unit (Pairwise-GRU) approach to analyze the collected time series data at each node. By considering the influence from the neighboring nodes in addition to historical data, we improve both the accuracy and confidence level of the prediction.

To analyze infrastructure network connectivity, we propose the probability propagation method (PrPm) and directed probability propagation method (dPrPm). These methods originate from the idea of belief propagation in graphical models, instead propagating a probability distribution to result in reliability assessments at all terminal nodes in the network. Both PrPm and dPrPm work for multiple-source networks with significantly reduced time complexity in computational efficiency compared to existing approaches. PrPm applies to general networks with approximated solutions. dPrPm provides upper bounds and lower bounds for acyclic directed networks.

To analyze infrastructure flow capacity, we propose algorithms to conduct multistate Bayesian network (BN) inference and a modified theory of maximum flow to deal with the multiple-source-to-multiple-sink scenario. In the multistate BN inference, we tackle the challenge of computational intractability for multistate BN by introducing a compression algorithm and corresponding inference algorithm for the network information. In the modified maximum flow work, we extend the traditional single-source-to-single-sink theory of maximum flow to the multiple-source-to-multiple-sink scenario with the ability to implement constraints and limit the feasible routes across the network. 

In sum, we present new methods to investigate the reliability of infrastructure flow networks in terms of connectivity and flow capacity. We include a Pairwise-GRU approach to improve the credibility of the inputs to the network. Two different methods are then proposed for both connectivity and flow capacity analysis. Over a range of applicable scenarios and networks, the methods are shown to advance current reliability assessment capabilities with increased accuracy and computational efficiency. 
 

Event Details

Date/Time:

  • Monday, April 26, 2021
    12:00 pm - 2:00 pm
Location: Atlanta, GA; REMOTE

Accessibility Information

Per accessibility compliance standards, this page may have links to files that would require the downloading of additional software:

  • Click here to download Microsoft Products.
  • Click here to download Adobe Reader.