PORPRS: Priority-aware task offloading in HAP-aided Internet of Vehicles via GRPO with Dynamic residual shrinkage networks

Abstract

In Internet of Vehicles (IoV), the uneven distribution of heterogeneous computing resources leads to unmet quality of service demands for tasks in certain areas. Leveraging their substantial computing resources, extensive spatial coverage, and the assistance of intelligent reflecting surfaces, high-altitude platforms present a viable solution to mitigate resource scarcity in remote IoV. Although existing research has significantly advanced the development of computation offloading strategies, many studies overlook the impact of task priority in heterogeneous scenarios on the overall quality of task execution. To address these problems, we propose a priority-aware computation offloading that evaluates task priority through a multi-dimensional framework. This framework considers delay constraints to ensure timely execution for delay-sensitive tasks, energy consumption to balance computing demands with system utility, regional computing resources to prioritize tasks in resource-constrained areas, and policy reliability to maintain the completion rate of tasks. To mitigate the impact of task priority on the overall quality of task execution, tasks with higher priority are executed first. Specifically, the computation offloading sequence is dynamically adjusted based on the task priority. On the basis of the dynamic sequence, to facilitate rational offloading strategy-making by agents, we design a group relative policy optimization with dynamic residual shrinkage networks, which enhances algorithm robustness by eliminating redundant features. Finally, extensive experiments are conducted on real datasets. The experimental results show that our algorithm can improve the completion rate of tasks while reducing delay and energy consumption.