As the title suggests, this dissertation is composed of three major topics. The first two are optimization problems focusing on developing effective solution methodologies, while for the last topic we present a large-scale information retrieval system in the domain of sports. With the algorithms and frameworks developed in this dissertation, we achieve a significant advancement in solving large-scale optimization problems with big data techniques, and in multiagent sports play retrieval via distributed computing and learning to rank. The operations in electric power control centers play a crucial role in ensuring the integrity of the nation’s electric grid and present formidable challenges to human operators. One of the primary challenges of information display for a power transmission control room application is the clutter from displaying too much information in too small of a display space. Thus, the task to optimize the layout of visual elements consisting of substations and transmission lines on the display interface is of vital importance. To this end, algorithms using several optimization techniques including Lagrangian relaxation and progressive hedging (PH) are proposed in the second chapter to make the interface less cluttered subject to human perceptual and cognitive capabilities. We conduct extensive computational studies to evaluate and compare the developed algorithms, and report our findings based on a real-world power grid in the U.S. We then move on to the airline fleet assignment problem. Since in recent years airlines around the world continue to face increasing capital and operational costs, they have been forced to cut costs and uphold revenue by utilizing their equipment capacity more efficiently to accommodate passenger demand. This is generally known as the fleet assignment problem, which deals with assigning aircraft of different capacities to the scheduled fight legs based on availabilities, costs, potential revenue and itinerary-based passenger demand. In order to address the high level of uncertainty in the market demand when the fleeting decisions are made and to capture the network effects (i.e., spill and recapture) for a more accurate estimate of passenger flow, we present a two-stage stochastic model which incorporates an attractiveness-based spill and recapture framework in the third chapter. This model considers spill and recapture based on passenger utility from itineraries. Solution approaches based on a distributed framework, namely MapReduce, are developed to reduce the computational time, and numerical results are reported using real data from a medium-size airline to evaluate and compare the proposed procedures. Finally, in the last chapter, we showcase a streaming-based retrieval system that can quickly find the most relevant basketball plays given an input query. The idea was inspired by the recent explosive growth of sports tracking data and the more and more important role sports analytics is playing in professional leagues. To search through a large number of games at an interactive speed, our system is built upon a distributed framework so that each query-result pair is evaluated in parallel. We also propose a pairwise learning to rank approach to improve search ranking based on users’ clickthrough behavior. The similarity metric in training the rank function is based on automatically learnt features from a convolutional neural network (CNN)-based autoencoder. In the end, we validate the effectiveness of our learning to rank approach by demonstrating rank quality in a user study.

Last modified

• 03/13/2018

Creator

• Mingyang Di

DOI

• https://doi.org/10.21985/N2WH5N

Subject

• Industrial Engineering and Management Sciences

Keyword

• Operational research
• Industrial Engineering

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright