移动边缘网络中基于强化学习的部分卸载分析

Research on partial offloading based on reinforcement learning in mobile edge networks

为解决边缘计算环境中多用户卸载问题,提出了一种基于强化学习的部分任务卸载方法,将二进制计算卸载扩展到连续动作域,从而更好地达到最小的时延,即采用Q-learning算法进行离散卸载决策,并基于深度确定性策略梯度算法(deep deterministic policy gradient algorithm for partial-offloading,DDPG-PO)连续卸载决策.实验结果表明所上述方法相较现有方法,提高了决策的准确性和实效性,能更好地降低时延.

In order to solve the problem of multi-user offloading in edge computing environment,a partial task offloading method based on Reinforcement Learning is proposed,which extends the binary computing offloading to the continuous action domain,so as to better achieve the minimum latency.The Q-learning algorithm is used to make discrete unloading decisions,and the DDPG-PO(deep deterministic policy gradient algorithm for partial-offloading,DDPG-PO)is used to make continuous offloading decisions.The experimental results show that the above methods improve the accuracy and effectiveness of decision-making compared to existing methods,and can better reduce latency.