Learning-based Resource Management for Internet of Vehicles

车辆互联网（IoV）已经被设想为通过提供许多新技术来实现下一代移动的网络，其是指使用V2 V（车辆到车辆）、V2 I（车辆到基础设施）、V2 H（车辆到人）和V2 S（车辆到传感器）交互的动态移动的通信系统。尽管在车联网的资源管理方面进行了大量的研究工作，但仍然缺乏支持高置信度车联网的实时应用的成熟科学，传统的分析工具/算法无法科普车联网的全部复杂性或充分预测系统行为，这是由于这些系统的高移动性，动态变化的环境和内在异构性带来的巨大挑战。这项研究的长期目标是推进基于学习的模型和算法的开发，用于在具有实时和低复杂度虚拟化资源管理的IoV网络中分配异构资源。最终，所有的研究成果和成果都将被整合到物理车联网网络和智能互联车辆系统中，以进一步推进此类系统的工程设计。 未来五年的短期目标有三个方面：1）建模和描述车辆移动性预测对资源管理的影响; 2）开发一个严格和系统的框架，用于设计基于动态学习的资源管理策略，这些策略可以在吞吐量最优、准时保证和计算复杂性三个维度上都取得出色的性能; 3）设计有效的资源分配算法，考虑各种工程成本和约束条件。拟议的研究包括三个信任：（1）小区内资源管理：每个智能小小区基站（SBS）将频谱资源分配给周围的车辆，旨在基于由机器学习驱动的流量预测模型最大化每个小区的吞吐量/效用;（2）全网资源管理：将通过考虑业务和频谱接入动态来开发车联网网络中SBS之间的自适应和最佳协作资源管理策略;（3）应用级资源管理：为了在关键性能指标、应用数据交付、服务质量和体验质量方面进一步提高系统性能，将开发具有网络切片的基于学习的应用级资源管理策略，以支持车辆的实时服务。预计拟议的研究计划将产生重大的科学，技术和社会影响，为支持车联网中的真实的实时应用提供科学研究基础。该计划还将有助于培养高素质的人才，为学员提供车联网研发方面的扎实知识和研究背景，使他们能够为加拿大高科技产业做出贡献。

Internet of Vehicles (IoV) has been envisioned to enable next--generation mobile networks by offering many new technologies, which refers to dynamic mobile communication systems using V2V (vehicle--to-vehicle), V2I (vehicle-to--infrastructure), V2H (vehicle--to--human) and V2S (vehicle-to-sensor) interactions. Despite significant research efforts on resource management for IoV, a mature science to support real-time applications of high--confidence IoV is still missing, and traditional analysis tools/algorithms are unable to cope with the full complexity of IoV or adequately predict system behavior, due to great challenges that arise from the high mobility, dynamic changing environment and intrinsic heterogeneity of such systems. The long-term goal of this research is to advance the development of learning-based models and algorithms for allocating heterogeneous resources in IoV networks with real-time and low-complexity virtualized resource management. Ultimately, all research contributions and outcomes will be integrated into physical IoV networks and smart connected vehicle systems to further advance the engineering of such systems. The short-term objectives in the next five years are three-fold: 1) To model and characterize the impact of vehicle mobility prediction on resource management; 2) To develop a rigorous and systematic framework for designing dynamic learning-based resource management policies that can achieve outstanding performance in all the three dimensions of throughput optimality, on-time guarantee, and computational complexity; 3) To design efficient resource allocation algorithms considering various engineering costs and constraints. The proposed research includes three trusts: (1)Intra-cell resource management: each smart small-cell base station (SBS) allocates the spectrum resource to the surrounding vehicles, aiming at maximizing the throughput/utility of each cell based on the traffic prediction model powered by machine learning; (2)Network-wide resource management: adaptive and optimal cooperative resource management policies among SBSs in IoV networks will be developed by considering both traffic and spectrum access dynamics; (3)Application-level resource management:  to further improve the system performance in terms of key performance indicators, application data delivery, quality of service, and quality of experience, learning-based application-level resource management policies with network slicing will be developed in favor of real-time services for vehicles. It is anticipated that the proposed research program will generate significant scientific, technological, and social impacts in providing the scientific research foundation for supporting real--time applications in IoV. This program will also contribute to the training of highly qualified personnel, providing the trainees with solid knowledge and research background in R&D of IoV, and enabling them to contribute to Canadian high technology industry.