Optimization of shared on-demand transportation

Abstract

Nations worldwide are experiencing significant urban growth, with over half of the global population currently residing in cities. This urbanization has increased urban commuting, leading to issues like congestion, air and noise pollution, and threats to public health. Recent years have witnessed a transformation in transportation driven by information technology, introducing innovative mobility solutions to tackle urban mobility challenges. These innovations encompass on-demand and shared mobility services, enhancing transportation efficiency and convenience. The integration of these solutions with public transit holds the potential to revolutionize the entire transportation system. Overcoming these challenges requires a comprehensive evaluation of the advantages and disadvantages of shared mobility services, ensuring a shift toward sustainable mobility in the future. This thesis aims to explore the optimization of on-demand transportation services in urban areas by employing methods in three aspects. The first part of this thesis analyzes historical travel time data from on-demand transport services, like taxis, to gain insights into traffic patterns and estimate arterial travel time precisely. It introduces a novel methodology that uses sparse GPS probe data and considers spatial correlations between network links. This research demonstrates the improved accuracy of travel time estimation by factoring in progressive spatial correlations. A case study in a partial network of New York City, using taxi data, shows enhanced travel time estimation accuracy, benefiting urban traffic optimization and congestion identification. The second part of this thesis centers on optimizing on-demand services with a real-time shuttle ridesharing algorithm. This novel algorithm efficiently matches ride requests to a fleet of vehicles, using a flexible simulation framework that adapts to different scenarios and incorporates real-time traffic data. By focusing on fleet capacities and tolerance times, the study shows that a reduced number of high-capacity taxis, along with optimized operational policies, significantly reduces waiting times and in-car delays for Manhattan taxi rides. The final part of this thesis focuses on developing precise short-term demand forecasting models for on-demand services, with an emphasis on deep learning techniques. It seeks to enhance prediction accuracy, investigate data granularity's impact, explore temporal and spatiotemporal variables, compare the model's performance with traditional and complex machine learning methods, and highlight the benefits of spatiotemporal considerations and vector embedding for improved prediction accuracy. The research presented in this thesis offers valuable implications for both research and practical applications. First, accurate estimates and predictions of travel times for urban links are crucial for optimizing urban traffic operations and identifying traffic congestion points. Providing precise travel time information offers benefits to users and operators by enabling them to choose better paths within the network and reduce overall travel time. Second, the potential of ridesharing services, optimized in real-time with dynamic traffic data, is shown by the proposed modular framework, together with the novel matching algorithm in Chapter 3 . The importance of system parameters and tailored operational policies to improve urban transportation systems gives valuable insight for the design of such services for operators. Moreover, the precise demand prediction can help the operators plan the fleet dispatching more efficiently.