基于深度强化学习的边缘计算资源分配方法

Edge Computing Resource Allocation Method Based on Deep Reinforcement Learning

边缘计算的特点使其具有广阔的军事应用前景。在边缘计算中引入联邦学习(federated learning,FL),考虑到物联网设备的资源有限,需要兼顾FL精度和设备能耗。提出了结合深度强化学习、联邦学习及自注意力机制的框架(DRL-FLSL)来实现选择设备并为其分配资源,目标是平衡FL精度和设备的能耗。该框架引入LSTM(long short term memory)预测网络状态,并添加多头自注意力机制实现更精准的信息提取。仿真实验结果表明,DRL-FLSL具有较好的训练效果,能够有效平衡FL精度和设备能耗。

The characteristics of edge computing make it have broad military application prospects.FL(Federated Learning)is introduced into edge computing.Considering the limited resources of IoT devices,FL accuracy and device energy consumption need to be taken into account.A framework combin⁃ing deep reinforcement learning,federated learning,and self attention mechanism(DRL-FLSL)is pro⁃posed to select devices and allocate resources to them,with the goal of balancing FL accuracy and device energy consumption.This framework introduces LSTM(Long Short Term Memory)to predict network state and adds a multi head self attention mechanism for more accurate information extraction.The simulation experimental results show that DRL-FLSL has super training effects and can effectively balance FL accu⁃racy and equipment energy consumption.