Adaptive UAV-Aided Fair Task Offloading with Fuzzy Caching for Vehicular Edge Computing via Hierarchical Deep Reinforcement Learning

초록

Vehicular edge computing (VEC) faces critical challenges due to overloaded ground networks, high traffic demand, and limited resources, resulting in delayed responses and task failures. To address these issues, we propose an adaptive ground-air collaboration that integrates roadside unit (RSU) and unmanned aerial vehicles (UAVs) networks. Unlike existing methods with a fixed UAV configuration, our system dynamically learns when to deploy or retract UAVs based on real-time conditions, thereby improving aerial resource utilization. We introduce UDMP-CTOP, a two-timescale hierarchical framework based on proximal policy optimization (PPO), which consists of UAV deployment management (UDMP) and collaborative task offloading (CTOP). A multi-objective utility function is designed to minimize system costs while ensuring regional fairness, preventing localized performance degradation during global optimization. In addition to learning optimal policies for task offloading and cache placement, we address the underexplored cache replacement problem under edge memory constraints. To this end, we propose a fuzzy logic-based cache replacement (FCR) mechanism that considers data recency, frequency, and volume. Extensive simulations demonstrate that the proposed algorithms achieve superior task success rates and low system costs under diverse conditions, yielding higher total utility across various topologies while reducing UAV energy consumption. © 1967-2012 IEEE.

키워드