Abstract
With the rapid development of global urbanization, the demand for efficient, safe, and environmentally friendly transportation systems is escalating. Vehicle-to-Everything (V2X) technology has emerged as a pivotal element in constructing intelligent transportation systems, offering innovative solutions to enhance road safety, alleviate traffic congestion, and mitigate environmental pollution through efficient communication between vehicles and infrastructure. In this context, the Roadside Units (RSUs) serve as a critical component of V2X technology, handling essential functions such as information collection, processing, and distribution, and are indispensable for the smooth operation of intelligent transportation systems. However, the deployment and operation of RSUs present financial considerations, including procurement, installation, and maintenance costs. Optimizing RSU deployment and enhancing its operational efficiency is thus crucial for achieving cost-effectiveness. Moreover, effectively leveraging the vast data collected by RSUs to extract valuable insights is a significant technical challenge in improving the utility value of RSU deployments. Addressing these challenges is vital for advancing the development of intelligent transportation systems.This dissertation conducts in-depth research on enhancing the utility effectiveness of RSUs, focusing on two primary research issues: 1) the optimal management of the large-scale deployment of RSUs to meet complex and diverse requirements and 2) the vehicle trajectory recovery problem based on RSU sensing data. These issues explore RSU deployment strategies and the application of RSU sensing data for trajectory recovery, respectively. The critical research content and innovations of this dissertation are as follows:
General RSU Deployment Algorithm for Complex Requirements. Aiming at the complexity, variability, and diversity that usually characterize RSU deployment requirements in V2X scenarios, this study proposes a generic RSU deployment algorithm to comprehensively evaluate deployment sites from a holistic level in a data-driven manner to satisfy diversified and comprehensive requirements. The algorithm adopts a Deep Reinforcement Learning (DRL) approach to model the RSU deployment problem as a Markov Decision Process and designs a feature analyzer to handle multimodal data and provide state representation for the DRL model. In addition, DRL components, including state representations, reward functions, and deployment actions, are customized for the RSU deployment problem, and an action space pruning scheme is proposed to accelerate model convergence. Evaluations on a real-world dataset show that the proposed algorithm in this study achieves significant performance improvement while being generalizable.
Multi-Granularity Trajectory Recovery Based on Roadside Units Sensing. This study proposes a new scenario for trajectory recovery based on RSU sensing data. Facing the problem of sparse data and uneven distribution in this scenario, a multi-granularity trajectory recovery algorithm that fully utilizes road information for enhancement is proposed. The algorithm realizes high-precision recovery of intersection-level and GPS-level trajectories based on graph neural network and Transformer algorithm by collaboratively utilizing the sensing data generated by vehicles passing through multiple RSUs. Experimental results show that the algorithm outperforms existing techniques in recovering missing trajectories outside the coverage of RSUs, providing an effective new solution for trajectory recovery in intelligent transportation systems.
| Date of Award | 28 Feb 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Shengjie Zhao (External Supervisor) & Zhenjiang LI (Supervisor) |