Privacy-Aware Task Offloading in Vehicular Edge Computing Based on Deep Reinforcement Learning

Abstract

Vehicular edge computing (VEC) faces critical challenges regarding limited onboard computational capacity and the potential for location privacy leakage. To address these issues, this paper proposes a novel, privacy-aware task offloading framework driven by deep reinforcement learning (DRL). First, we design a dynamic caching mechanism at roadside units (RSUs) to store previously computed task outputs, reducing redundant computation and communication overhead. Second, we introduce privacy entropy to quantitatively measure and enforce location privacy protection, mitigating the risk of RSU-induced data breaches. We then formulate a joint optimization problem that balances latency, energy consumption, and privacy preservation. To efficiently solve this, a deep deterministic policy gradient algorithm enhanced with prioritized experience replay (PER-DDPG) is developed to learn the optimal offloading strategy. Simulation results demonstrate that our proposed approach outperforms existing benchmark schemes, delivering superior convergence speed, lower latency, higher energy efficiency, higher privacy entropy and an improved task success rate.