移动边缘计算中基于深度强化学习的依赖任务卸载研究

Dependent Task Offload Based on Deep Reinforcement Learning in Mobile Edge Computing

在移动边缘计算中,移动设备将计算密集型应用卸载到边缘服务器上运行,以提高服务质量和保护数据安全.应用通常由具有依赖关系的任务组成,传统方法在解决依赖任务卸载问题时效率低下并且忽略了信道动态变化对卸载策略的影响.本文通过建立信道状态变化场景下的依赖任务卸载模型,以最小化应用的响应时间和移动设备能耗为优化目标,提出一种依赖任务卸载算法(Dependent Task Offloading Algorithm,DTOA),将依赖任务卸载转化为马尔可夫决策过程下的最优策略问题,基于优先经验回放的DDQN求解获得有效的任务卸载策略.实验结果表明,对于不同依赖关系的任务,与已有卸载方法相比,本文提出的策略有效降低了应用的响应时间和移动设备的能耗.

In mobile edge computing platforms,mobile devices offload compute-intensive applications to edge servers for computing to improve the quality of service and protect data security.An application usually consists of tasks with dependencies,and traditional optimization methods are inefficient and ignoring the impact of channel dynamics on the offloading strategy.This paper proposes a Dependent Task Offloading Algorithm(DTOA) to minimize application response time and mobile device energy consumption by establishing a dependent task offloading model in the scenario of channel state change.The task offloading problem is transformed into an optimal strategy under Markov decision process,and offloading strategy is obtained based on DDQN solution with priority experience replay.Compared with existing offloading methods,the proposed strategy effectively reduces application response time and mobile device power consumption.