基于学习博弈的D2D任务卸载激励机制  

Learning game based incentives for D2D task offloading

作　　者：任则鹏 黎作鹏 琚恒帅 REN Ze-peng;LI Zuo-peng;JU Heng-shuai(Department of Information and Electrical Engineering,Hebei University of Engineering,Handan 056038,China)

机构地区：[1]河北工程大学信息与电气工程学院,河北邯郸056038

出　　处：《计算机工程与设计》2025年第2期438-446,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(62101174);河北省自然科学基金项目(F2021402005)。

摘　　要：为解决任务设备卸载决策问题并激励空闲设备加入任务卸载,将卸载决策问题转换为多对一匹配博弈,采用盖尔-沙普利算法实现稳定匹配;提出一种基于斯坦克伯格博弈的激励机制,在信息完全共享的前提下证明了SE(Stacklberg Equilibrium,SE)的存在性和唯一性。在保护用户隐私的前提下,提出一种多维动作的深度强化学习算法,每个博弈参与方都是一个智能体,通过参考其它人的历史策略来学习最优策略。通过仿真实验验证了所提算法的有效性。To solve the task device offloading decision problem and motivate idle devices to join the task offloading,the offloading decision problem was converted into a many-to-one matching game and stabilized matching was achieved using the Gale-Shapley algorithm.An incentive mechanism based on Stacklberg game was proposed,and the existence and uniqueness of SE(Stacklberg Equilibrium,SE)was proved on the premise of complete information sharing.On the premise of protecting user privacy,a deep reinforcement learning algorithm for multidimensional actions was proposed,where each game participant was an intelligent body that learned the optimal strategy by referring to the historical strategies of others.The effectiveness of the proposed algorithm is verified by simulation experiments.

关 键 词：移动边缘计算 端到端 斯坦克伯格博弈 强化学习 激励机制 任务卸载 匹配博弈 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]