基于改进SMDP的车载任务卸载决策算法

针对边缘服务器的负载过重问题,可以将路边空闲车辆以及移动车辆应用虚拟化技术整合成资源池,为时延敏感类任务提供弹性服务；由此建立了一个分组传输的通信系统模型,为降低二进制指数退避算法中的信道碰撞概率,采用基于网络车辆节点的数量来适当调整最小竞争窗口的方法；结合分配资源的时序决策特点,提出车载边缘计算系统中基于改进的半马尔科夫决策过程的计算卸载策略,在制定系统动作的最优策略时,引入带有余弦项的非线性权重因子,对立即收益和未来期望收益进行动态加权,根据贝尔曼方程进行价值迭代,实现系统长期收益的最大化；仿真结果表明,所提策略能有效降低卸载时延,提高系统吞吐量,同时系统的长期收益也有显著的提升。

An Offloading Algorithm of Vehicle Tasks Based on Improved SMDP

To solve the overload problem of edge servers, the virtualization technology can be applied to integrate idle vehicles on the roadside and mobile vehicles into a resource pool to provide elastic services for delay-sensitive tasks. Therefore, a communication system model of packet transmission is established. In order to reduce the probability of channel collision in the binary exponential backoff algorithm (BEB), the minimum competition window is adjusted according to the number of vehicle nodes in the network. Considered the characteristics of timing decision of resource allocation, a computational offloading strategy is proposed for vehicle edge computing (VEC) system based on the improved semi-Markov decision process(SMDP). When formulating the optimal strategy of system action, nonlinear weight factor with cosine term is introduced to dynamically weight immediate reward and future expected reward. An iterative algorithm based on Bellman equation is utilized to approach the maximum long-term reward. Simulation results show that the proposed strategy can effectively reduce offloading delay, optimize throughput, and significantly improve the long-term rewards of the system.