Smart Sustainable Mobility: Analytics and Algorithms for Next-Generation Mobility Systems

To this date, mobility ecosystems around the world operate on an uncoordinated, inefficient and unsustainable basis. Yet, many technology-enabled solutions that have the potential to remedy these societal negatives are already at our disposal or just around the corner. Innovations in vehicle technology, IoT devices, mobile connectivity and AI-powered information systems are expected to bring about a mobility system that is connected, autonomous, shared and electric (CASE). In order to fully leverage the sustainability opportunities afforded by CASE, system-level coordination and management approaches are needed. This Thesis sets out an agenda for Information Systems research to shape the future of CASE mobility through data, analytics and algorithms (Chapter 1). Drawing on causal inference, (spatial) machine learning, mathematical programming and reinforcement learning, three concrete contributions toward this agenda are developed. Chapter 2 demonstrates the potential of pervasive and inexpensive sensor technology for policy analysis. Connected sensing devices have significantly reduced the cost and complexity of acquiring high-resolution, high-frequency data in the physical world. This affords researchers the opportunity to track temporal and spatial patterns of offline phenomena. Drawing on a case from the bikesharing sector, we demonstrate how geo-tagged IoT data streams can be used for tracing out highly localized causal effects of large-scale mobility policy interventions while offering actionable insights for policy makers and practitioners. Chapter 3 sets out a solution approach to a novel decision problem faced by operators of shared mobility fleets: allocating vehicle inventory optimally across a network when competition is present. The proposed three-stage model combines real-time data analytics, machine learning and mixed integer non-linear programming into an integrated framework. It provides operational decision support for fleet managers in contested shared mobility markets by generating optimal vehicle re-positioning schedules in real time. Chapter 4 proposes a method for leveraging data-driven digital twin (DT) frameworks for large multi-stage stochastic design problems. Such problem classes are notoriously difficult to solve with traditional stochastic optimization. Drawing on the case of Electric Vehicle Charging Hubs (EVCHs), we show how high-fidelity, data-driven DT simulation environments fused with reinforcement learning (DT-RL) can achieve (close-to) arbitrary scalability and high modeling flexibility. In benchmark experiments we demonstrate that DT-RL-derived designs result in superior cost and service-level performance under real-world operating conditions