Thesis Title: Demand Projection and Complex Resource Allocation Decisions on Networks
Thesis Committee:
Dr. Pinar Keskinocak (advisor), School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Mohit Singh (advisor), School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Lauren Steimle, School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Mathieu Dahan, School of Industrial and Systems Engineering, Georgia Institute of Technology
Dr. Ozlem Ergun, Mechanical and Industrial Engineering, Northeastern University
Date and Time: Wednesday, August 2, 2023, 12:00 - 1:30 pm ET
In-person location: ISyE Main 228
Online Meeting Link: Teams
Abstract:
With the increase in human-to-nature and human-to-human interactions (e.g., urbanization, exploitation of the natural environment, and globalization), the risk and the impact of global pandemics (e.g., SARS, Ebola, COVID-19) or large-scale disasters (e.g., earthquakes, hurricanes, refugee crises) are heightened. Therefore, responding to health and humanitarian needs by projecting demands and efficiently allocating limited resources in the event of a global pandemic or disaster is of paramount importance.
In response to a global pandemic (such as Covid-19), it is essential to model the infectious disease, project the spread and need for healthcare resources geographically and over time, and assess the impact of non-pharmaceutical intervention strategies. Therefore, we utilize a detailed agent-based simulation model that captures the disease progression in an individual (natural history) and the spread in the population through a contact network (representing interactions in households, peer groups, such as schools and workplaces, community, travel between geographic locations, etc.)
In Chapter 2, we evaluate the effectiveness and impact of non-pharmaceutical intervention decisions, including school closure, shelter-in-place, and voluntary quarantine, using an agent-based simulation model in the state of Georgia. We test various scenarios with different shelter-in-place duration and time-varying compliance levels to voluntary quarantine to inform decision-makers about potential social distancing recommendations to be shared with the public. Our analysis shows that shelter-in-place followed by voluntary quarantine could substantially reduce COVID-19 infections, healthcare resource needs, and severe outcomes, delay the peak, and enable better preparedness.
In Chapter 3, we quantify the impact of non-pharmaceutical interventions and the trade-offs between the potential benefits (e.g., reduction in infection spread and adverse outcomes) and economic consequences (i.e., refraining from workplace and community), as measured by the number of homebound people or person-days. Such evaluations can assist local and national decision-makers in choosing different combinations of targeted interventions over time to reduce infection spread while considering the societal and economic impact. Our findings highlight that targeted interventions such as voluntary quarantine or voluntary quarantine combined with school closure significantly reduce the infection spread without causing social and economic disruption. On the other hand, large-scale interventions like shelter-in-place temporarily slow down the infection spread but are highly disruptive to society.
Disaster debris management includes debris clearance from roads to provide access to critical services, and it is followed by debris collection, i.e., collecting and removing debris from the affected area. The overwhelming nature of disaster debris collection demands resources far exceeding municipalities’ abilities to provide additional labor, equipment, and services. Hence, municipalities engage multiple contractors and face with the challenging task of partitioning the affected area into zones and assigning each zone to a contractor to collect and dispose of the debris in a timely manner. Motivated by this, we consider the problem of partitioning a weighted graph into connected subgraphs, where weights correspond to the processing requirements of vertices, and each subgraph is assigned to (and processed by) a resource. The resources are heterogeneous, where each resource has a speed (processing capacity); the completion time of a subgraph/resource is node- and resource-specific (i.e., processing times are resource-dependent). The objective is to minimize the makespan, i.e., the maximum completion time across all resources.
In Chapter 4, we introduce the problem and thoroughly study the optimization landscape of the problem, both from a theoretical perspective as well as heuristic methods that solve the problem at scale. We show that the problem is NP-complete and classify many special cases as NP-complete. We formulate this problem as integer and mixed-integer programs; we propose approximation algorithms with provable bounds for some cases and provide several heuristics. We present results of an extensive computational study, including a realistic hurricane scenario for the state of Florida, which is generated using FEMA’s hazard estimation tool Hazus.
In Chapter 5, we focus on special cases of the problem where we consider underlying graph structures such as connectivity and planarity and explore how it impacts the computability of the problem. We show that the decision version of the problem remains NP-complete even when the input graph has these special structures. We propose approximation algorithms with theoretical worst-case bounds on the performance of the algorithm for special graph structures and specific numbers of resources. Considered special graph structures include k-connected graphs, where k = 2; 3; 4, and min{k;4}-connected planar graphs, where k corresponds to the number of resources. Building on these structural insights, we also propose heuristics that work on graphs where such structures (i.e., connectivity) might not be present. We present the results of a computational study focusing on a hurricane scenario for the state of Florida.