面向车联网的多智能体强化学习边云协同卸载

车联网边缘计算是实现车联网系统低时延和高可靠性的关键技术，但现有方法普遍存在场景趋同和系统建模局限的问题，同时包含复杂的训练过程并面临维灾风险。通过结合云计算技术，提出一种基于多智能体强化学习的边云协同卸载方案。依据随机几何理论计算卸载节点覆盖概率，对车辆节点与卸载对象进行预配对。利用线性Q函数分解方法反映每个智能体多效用因子与任务决策间的映射关系，通过云端协同机制将智能体决策记录作为经验上传到云端，并在云端将训练更完备的神经网络反馈到边缘节点。仿真结果表明，该方案在功耗和延时方面性能优于单一固定边缘的计算策略，且算法复杂度较低，能够有效提升边云协同卸载能力，实现低时延、高可靠的任务卸载。

Collaborative Edge and Cloud Offloading for Internet of Vehicles Using Multi-Agent Reinforcement Learning

Edge computing for Internet of Vehicles(IoV)is key to realizing highly reliable and low-latency IoV systems.However,existing methods generally have the problems of scene convergence and system modeling limitations,and are faced with complex training processes and disaster maintenance risks.By combining the cloud computing technology,this paper proposes a collaborative edge and cloud offloading scheme based on multi-agent reinforcement learning.The strategy uses the stochastic geometry theory to calculate the coverage probability of the offloading nodes and pre-match the vehicular nodes to offloading objects.On this basis,the linear Q function decomposition method is used to reflect the mapping relationship between each agent’s multi-utility factor and task decision.Then through the collaborative cloud and edge computing mech anism,each agent’s decision records are uploaded to the cloud as experience,and the more comprehensively trained neural network is returned to the edge nodes.The results of simulation show that the proposed scheme outperforms the computing strategies using only fixed edge servers in terms of power consumption and latency.The method reduces the algorithm complexity,and can significantly improve the collaborative edge and cloud offloading ability to realize highly reliable and low-latency task offloading.