Mobility-Prediction based Proactive Edge Caching in Vehicular Networks

This thesis studies and designs machine intelligence based mobility-prediction algorithms to address the proactive edge caching problem in vehicular networks. In particular, the thesis focuses on predicting the next road-side unit (RSU) along a vehicle's route as the node for proactively caching content and meanwhile investigates approaches to improve such prediction accuracy. Firstly, the thesis presents an offline sequence prediction based proactive caching (SPPC) system that employs Compact Prediction Tree+ (CPT+) algorithm by modelling RSUs as symbols of sequences. The proposed system explores the feasibility and the performance of applying the next RSU prediction to proactive caching. Moreover, to achieve better adaptability of learning approaches, the thesis then proposes an online bandit learning approach by designing two novel multi-armed bandit (MAB) based proactive caching systems. They are proven to have better prediction accuracy than other comparative systems, and hence better network performance in terms of average network delay. In addition, the MAB-based learning systems are also evaluated with an extended uncertainty analysis framework, Subjective Logic, using entropy, and they demonstrate improved uncertainty reduction during learning, which shows analytical evidence of their better prediction accuracy. Furthermore, with the aim of fully exploring the potential of cMAB learning, the thesis proposes a Hybrid cMAB Proactive Caching (HCPC) system which implements Dual-context cMAB and Single-context cMAB algorithms and is further developed into two system variants: Vehicle-Centric and RSU-Centric. They allow RSUs to adaptively finalise their predictions by a specific switching mechanism and improve the prediction accuracy to a new level. Lastly, to verify the applicability and adaptability of the proposed algorithms and systems in this thesis, traffic simulation with Simulator of Urban MObility (SUMO) in two major cities, Las Vegas, USA and Manchester, UK, with different road layouts, is performed and various traffic scenarios are tested and assessed.