Mining Heterogeneous Spatial-Temporal Data with Graph Neural Network to Support Smart City Management
Publisher
The University of Arizona.Rights
Copyright © is held by the author. Digital access to this material is made possible by the University Libraries, University of Arizona. Further transmission, reproduction, presentation (such as public display or performance) of protected items is prohibited except with permission of the author.Abstract
Daily life in urban areas is challenged by the increasing population of cities with limited resources and services. Widespread adoption of the Internet of Things, machine learning, and big data technology makes it possible to monitor and manage resources efficiently in smart cities. Smart city applications are trying their best efforts to combine and utilize all heterogeneous big data resources generated by IoT sensors to build management systems and provide intelligent and high-quality services in areas such as public transportation, sustainable resource management, and quality of life for users. An intelligent transportation system is a significant application in smart cities to provide innovative services and enable users to make safer and more convenient use of public transportation networks by storing, visualizing, and managing Spatial-temporal data to model and solve real-world public traffic problems. Towards this end, this dissertation aims to address the challenges in processing, representing, and mining large-scale heterogeneous Spatial-temporal data by collaborating key techniques like big data analytics, network analysis, and deep learning. More specifically, I first explore how to collect and process all possible heterogeneous public data that could help model Spatial-temporal data in the public transportation system. Then, I design a novel three-level prediction model to predict the pick-up and drop-off demand in a bike-sharing system combining machine learning and network analysis technologies. Finally, I extend the work to a general traffic flow prediction problem and demonstrate a deep learning graph neural network model to extract high-level data representations to improve prediction performance.The dissertation explores the general applications in smart city management system applications and focuses on the intelligent transportation system domain. For intelligent transportation systems, mobility prediction is important for providing high-quality services like reducing delays and waiting times, improving travel conditions, and ensuring a safe and clean environment. The bike-sharing system is introduced to public transportation systems as a solution for those purposes in recent decades and is becoming more and more popular in large smart cities. In the first essay, I explore the feature of Spatial-temporal data in public transportation systems and generate tools to help store, visualize and manage data in the public transportation system. In the second essay, I design a three-level prediction model to predict the whole system level bike demand, station level pick-up demand, station level drop-off demand in bike-sharing systems. Three-level prediction model could give the city manager insight to understand the overall usage trend and help to design a rebalancing strategy in the future based on station-level detailed prediction results. At last, the research problem is extended to a general inflow and outflow mobility problem in transportation. I design a graph neural network algorithm to better represent the dynamic Spatial-temporal features in real-world transportation and achieve higher prediction accuracy compared with the result from the three-level prediction model in the second essay. Through this dissertation, I show a general framework to manage and store data in public transportation in an efficient way. The bike-sharing system is the latest introduced alternative public transportation system and I propose different models to represent the features and predict the public mobility pattern based on the real-world heterogeneous dataset. The mobility pattern prediction result could be converted into decision-making actions for city managers in future intelligent transportation systems design such as traffic facility planning, traffic intervention design, and resource allocation design.Type
textElectronic Dissertation
Degree Name
Ph.D.Degree Level
doctoralDegree Program
Graduate CollegeManagement Information Systems