Sensing and Learning In Structured Non-stationary Environments
2024-01
Loading...
View/Download File
Persistent link to this item
Statistics
View StatisticsJournal Title
Journal ISSN
Volume Title
Title
Sensing and Learning In Structured Non-stationary Environments
Authors
Published Date
2024-01
Publisher
Type
Thesis or Dissertation
Abstract
In large infrastructure systems, sensing is key to maintain high quality information for decision making, while sometimes the non-stationary nature of the systems and short-lived agents in the systems make it hard to merge exploration and exploitation. One way to overcome such difficulty is to explore the resources in the system and information can be shared with all decision makers, in my dissertation, we’ll take the national airline as the examples to study the structured non-stationary environments. Aircraft in the National Aviation System (NAS) often rely on wind information from the National Oceanic and Atmospheric Administration (NOAA) to calculate favorable paths. However, wind conditions are dynamic and the NOAA information becomes quickly outdated because it is based upon sparse sampling both in space and time. This leads to inefficient, slower, paths used in practice. A goal of the Federal Aviation Administration’s (FAA) NextGen program is to use dynamic information to reduce inefficiencies. One such way to obtain high quality dynamic information and reduce inefficiency is to use en-route aircraft as ‘sensors’. This raises a natural question, “if a fraction of the aircraft can be used for sampling, how should aircraft be routed to provide most useful information for other aircraft to minimize system costs?” To answer this question, we begin with a stylized model of the aircraft routing problem, and capture the uniquely spatial and temporal correlations in wind dynamics. In Chapter 2, we analyzed the travel time at a paths' level, by modeling spatial and temporal correlation between the travel time along different paths, and formulating the travel time as a Brownian surface. Under this uncertainty structure, we address two questions: (i) if an offline schedule of paths to be sampled is desired, what is the optimal sampling schedule? and (ii) if the paths to be sampled are to be chosen in real time according to flight schedules, what is a near-optimal sampling policy? We provide answers to these questions using state-independent policies and state-dependent policies with provable guarantees in Chapter 3. In Chapter 4, we generate a comprehensive testbed from real-world flight data and computationally evaluate the performance of our sampling policies. Our testbed consists of seventeen origin-destination airport pairs, with five short-haul, seven medium-haul and five long-haul pairs. Our results show that collecting the right information and utilizing it to plan future aircraft routes could reduce a flight's travel time and associated fuel burn by 5% on average. Our modeling framework and results are also applicable to smaller, intra-city aircraft and unmanned aircraft such as UAVs and drones. The promising results and discoveries inspire us to delve deeper into the realm of bandit problems in structured non-stationary environments. After thoroughly surveying and summarizing the existing work, in Chapter 5, we extend the non-stationary structures to the combinatorial settings. This expansion allows us to examine how the intersection of links affects the common arcs in airline networks. Our goal is to explore the impact of link intersections on paths that share common arcs and to propose a near-optimal policy for this scenario. We approach this investigation from two directions, each offering potential solutions. We conclude in Chapter 5 with the discussion of the ongoing jobs and promising future research directions.
Keywords
Description
University of Minnesota Ph.D. dissertation. 2023. Major: Industrial and Systems Engineering. Advisor: Ankur Mani. 1 computer file (PDF); 125 pages.
Related to
Replaces
License
Collections
Series/Report Number
Funding information
Isbn identifier
Doi identifier
Previously Published Citation
Suggested citation
Gao, Jing. (2024). Sensing and Learning In Structured Non-stationary Environments. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/262863.
Content distributed via the University Digital Conservancy may be subject to additional license and use restrictions applied by the depositor. By using these files, users agree to the Terms of Use. Materials in the UDC may contain content that is disturbing and/or harmful. For more information, please see our statement on harmful content in digital repositories.