Organizational Unit:

H. Milton Stewart School of Industrial and Systems Engineering

Permanent Link

https://hdl.handle.net/1853/70758

Parent Organization

Organizational Unit

College of Engineering

Includes Organization(s)

Organizational Unit

William M. Keck Virtual Factory Lab (VFL)

ArchiveSpace Name Record

https://finding-aids.library.gatech.edu/agents/corporate_entities/1032

Full item page

Publication Search Results

Now showing 1 - 10 of 606

USING WIFI MOBILITY DATA FOR MODELING COVID-19 ON UNIVERSITY CAMPUSES

(Georgia Institute of Technology, 2021-05-05) Xie, Jiajia

Infectious diseases, like mumps, flu, or measles, can cause devastating impacts on universities. To protect the community's health, schools must learn how to operate during epidemics. In the light of Covid-19, for instance, the universities in the U.S. have struggled to bring students, staff, and faculties back to campuses. On the one hand, these schools are often hotspots for outbreaks. On the other hand, a long-term local lockdown will likely incur losses of financial income for the school and the local businesses due to diminishing enrollment and limited visits to campuses and their surrounding neighborhoods To meet the reopening goal, the schools must assess the ongoing epidemic of Covid-19 on campus and design operational plans with more robust and accurate information than the data provided by other local agencies as support. Frequently asked questions from the perspective of campus officials are: How can we predict potential outcomes of disease spread? How can we evaluate strategies to control the epidemic? Which groups of individuals and locations are particularly vulnerable to Covid-19? How can we prioritize the testing program among individuals active on campus? Answering those questions typically involves disease modeling since models help us abstract the disease dynamics and reason more about the mechanism of disease transmissions among the community. This thesis targets several natural and fundamental problems for universities during the Covid-19 pandemic using human mobility data. We propose using the on-campus WiFi infrastructure to understand human mobility and approximate contact networks among individuals on campus. When an individual accesses the WiFi on campus, their device sends a request to a WiFi access point which creates a record that the device was connected to the WiFi network. From these logs, we can determine when and for how long that individual was connected to the WiFi through a particular access point and infer the location of that individual to the level of a room on campus. More formally, the logs give us a bipartite network between users and WiFi access points across different time-stamps, defined as WiFi Mobility data. Each connection of a user to a log at any time will be recorded as an edge in the bipartite network. Using a projection of this bipartite network, we can infer which individuals come into close proximity of each other on campus. Each connection of a user to a log at any time will be recorded as an edge in the network. We construct and validate a network-based simulation model of Covid-19 on university campuses using Wifi mobility data to approximate the contact network among individuals. Then, we design and evaluate two novel methods for improving decision-making powered by the WiFi mobility data with the model constructed. The first method outputs a more granular and localized closure policy, causing more effective disease intervention outcomes but less burdensome to individuals and schools. The second can discover likely chains of transmission among individuals and missing infections on campus given the current testing report data.
Data-driven optimization for police beat design in South Fulton, Georgia

(Georgia Institute of Technology, 2020-04-28) Lu, Le

We redesign the police patrol beats in South Fulton, Georgia, in collaboration with the South Fulton Police Department (SFPD), using a predictive data-driven optimization approach. Due to rapid urban development and population growth, the original police beats arrangement designed in the 1970s was far from efficient, which leads to low policing efficiency and long 911 call response time. We balance the police workload among different regions in the city, improve operational efficiency, and reduce 911-call response time by redesigning beat boundaries for the SFPD. We discretize the city into small geographical atoms, which correspond to our decision variables; the decision is to map the atoms into "beats", which are the basic units of the police operation. We analyze workload and trend in each atom using the rich dataset for police incidents reports and U.S. census data and predict future police workload for each atom using spatial statistical regression models. Basing on this, we formulate the optimal beat design as a mixed-integer programming (MIP) program with contiguity and compactness constraints on the shape of the beats. The optimization problem is solved using simulated annealing due to its large-scale and non-convex nature. Our resulted beat design can reduce workload variance by over 90% according to our simulation. Our new optimal beat design has been approved by the City Council of South Fulton and implemented in January 2020.
Online sufficient dimensionality reduction for sequential high-dimensional time-series

(Georgia Institute of Technology, 2015-01-08) Li, Qingbin

In this thesis, we present Online Sufficient Dimensionality Reduction (OSDR) algorithm for real-time high-dimensional sequential data analysis.
Natural Gas Storage Valuation

(Georgia Institute of Technology, 2007-11-16) Li, Yun

In this thesis, one methodology for natural gas storage valuation is developed and two methodologies are improved. Then all of the three methodologies are applied to a storage contract. The first methodology is called "intrinsic rolling with spot and forward", which takes both the spot and forward prices into account in the valuation. This method is based on the trading strategy by which a trader locks the spot and forward positions by solving an optimization problem based on the market information on the first day. In the following days, the trader can obtain added value by adjusting the positions based on new market information. The storage value is the sum of the first day's value and the added values in the following days. The problem can be expressed by a Bellman equation and solved recursively. A crucial issue in the implementation is how to compute the expected value in the next period conditioned on the information in current period. One way to compute the expected value is Monte Carlo simulation with ordinary least square regression. However, if all of the state variables, spot, and forward prices are incorporated in the regression there are too many terms, and the regression becomes uncontrollable. To solve this issue, three risk factors are chosen by performing principle component analysis. Dimension of the regression is greatly reduced by only incorporating the three risk factors. Both the second methodology and the third methodology only consider the spot price in the valuation. The second methodology uses Monte Carlo simulation with ordinary least square regression, which is based on the work of Boogert and Jong (2006). The third methodology uses stochastic dual dynamic programming, which is based on the work of Bringedal (2003). However, both methodologies are improved to incorporate bid and ask prices. Price models are crucial for the valuation. Forward prices of each month are assumed to follow geometric Brownian motions. Future spot price is also assumed to follow a geometric Brownian motion but for a specific month its expectation is set to the corresponding forward price on the valuation date. Since the simulation of spot and forward prices is separated from the storage optimization, alternative spot and forward models can be used when necessary. The results show that the value of the storage contract estimated by the first methodology is close to the market value and the value estimated by the Financial Engineering Associates (FEA) provided function. A much higher value is obtained when only spot price is considered, since the high volatility of the spot curve makes frequent position change profitable. However in the reality traders adjust their positions less frequently.
Modeling expertise in the design of warehousing and distribution systems

(Georgia Institute of Technology, 2001-05) Zerangue, Natalie Frances
The VProf tutor : teaching MD-11 pilots vertical profile navigation

(Georgia Institute of Technology, 2001-05) Gray, William Michael
Investigating the role of procedures and cockpit display of traffic information in candidate air traffic management operations

(Georgia Institute of Technology, 2000-05) Yankosky, Leonard Joseph
Effects of displays and alerts on subject reactions to potential collisions during closely spaced parallel approaches

(Georgia Institute of Technology, 1998-12) Vandor, Balazs
Asymptotics of k-limited polling models

(Georgia Institute of Technology, 1998-12) Chang, Woojin
Use of optimization models to solve labor planning and scheduling problems for the service industry

(Georgia Institute of Technology, 1998-05) Summers, Deborah A.