|
چکیده
|
Vehicular Edge Computing (VEC) has emerged as a transformative paradigm in Intelligent Transportation Systems (ITS), enabling vehicles to offload computationally demanding and latency-sensitive tasks to nearby edge or cloud servers. However, efficient task offloading remains challenging due to the dynamic nature of vehicular networks, heterogeneous edge resources, variable service costs, and diverse task requirements. Traditional approaches often fail to distinguish between delay-sensitive and delay-tolerant tasks, leading to suboptimal trade-offs between response time and monetary cost. To address these limitations, this paper proposes PACOB, a priority-aware computation offloading framework based on a multi-armed bandit model enhanced with an Upper Confidence Bound (UCB) mechanism. Unlike conventional methods, PACOB employs dual action-value functions to explicitly distinguish task types and enable task-specific optimization. For delay-sensitive tasks, it prioritizes minimizing response time to meet strict deadlines, whereas for delay-tolerant tasks, it focuses on cost efficiency while preserving service quality. By dynamically learning optimal offloading policies and adapting to real-time network variations, PACOB achieves a balanced optimization of latency and cost across heterogeneous environments. Extensive simulations across diverse VEC scenarios demonstrate PACOB’s superiority over four baselines: Random (RND), Priority-Based Heuristic (PBH), -greedy, and standard UCB. Compared with standard UCB, PACOB reduces average response time by about 35% and improves the percentage of deadline-satisfied tasks (PDST) by 18%, with only a moderate 25% rise in cost for delay-sensitive tasks. For delay-tolerant tasks, PACOB and PBH show comparable cost efficiency, but PACOB achieves slightly better response time and PDST through adaptive redirection of some tasks to underutilized edge servers. These results confirm PACOB’s robustness and adaptability in balancing latency, reliability, and cost efficiency for real-time vehicular edge computing.
|