Energy Efficient Algorithm for Multi-User Adaptive Edge Computing Offloading in Vehicular Networks Based on Meta Reinforcement Learning

In the evolving Internet of Vehicles (IoV) landscape, the dual demands of low-latency and high computational power pose significant challenges, especially given the limited computational capabilities of internet-enabled autonomous vehicles and latency issues in cloud computing. This study proposes a novel, energy-efficient algorithm for multi-user adaptive edge computing offloading in vehicular networks, designed to optimize computational resource allocation. This reduces vehicle energy consumption while managing latency in a multi-user environment. At the heart of this strategy is a meta-reinforcement learning-based framework, adept at addressing the dynamic and complex nature of IoV environments. It uniquely integrates a directed acyclic graph with a model-independent meta-learning architecture, thereby enhancing the task offloading process's efficiency and adaptability. This allows the system to rapidly adjust to environmental changes with minimal data, facilitating the development and implementation of efficient offloading strategies for edge computing. These strategies are crucial in significantly reducing energy consumption and maintaining latency performance across various user scenarios. Empirical assessments of this algorithm highlight its effectiveness in reducing energy usage and optimizing training latency, effectively balancing the energy-latency trade-off in IoV applications. The results indicate a significant advancement in computational resource management for multi-user vehicular networks.