车联网环境下基于强化学习的边缘服务器部署策略

Edge server deployment strategy with reinforcement learning in Internet of vehicles

鉴于现有的边缘服务器部署策略主要用于改善5G、无线城域网等场景下的服务性能,无法直接用于车联网服务部署,提出一种边云协同的5G车联网边缘计算系统模型,针对该系统模型设计了基于强化学习的边缘服务器部署策略,其以负载优化为核心目标,在保证低延迟和低能耗前提下实现边缘服务器间的负载均衡。根据路边单元位置信息用Canopy聚类获取初始的聚簇数,用模糊C均值聚类获取路边单元的初始划分,并输出路边单元归属优先级矩阵;通过强化学习获得路边单元归属的最优状态并计算聚簇中心作为边缘服务器部署位置。通过对比实验验证了该策略在低服务延迟和低能耗下,能够高度实现边缘服务器间的负载均衡,表明该策略具有优越性。

The existing edge server placement methods are mainly used to improve service performance in scenarios such as 5 G and wireless metropolitan area networks, but cannot be directly used for the deployment of Internet of Vehicles(IoV) services. Therefore, an edge computing system model with edge-cloud collaboration for 5 G-IoV was proposed, and a deployment Strategy of edge servers based on Reinforcement Learning(SRL) was designed for this system model. Specifically, load optimization was taken as the core goal, and the load balancing among edge servers was realized under the premise of low delay and consumption. According to the location information of the roadside unit, the clustering algorithm Canopy was used to calculate the initial number of clusters. The initial division of the roadside unit was obtained using fuzzy C-means, and the roadside unit attribution priority matrix was output. Through the reinforcement learning, the optimal state of the roadside unit was obtained and the cluster center was calculated as the deployment location of the edge server. Comparative experiments verified that SRL had achieved a high degree of load balancing between edge servers under the premise of low service delay and consumption, which demonstrated the superiority of SRL.