基于无线射频供能的端边协同智能任务推理机制

Device-edge Collaborative Intelligent Task Inference Mechanism Based on Wireless RF Energy Harvesting

针对云侧进行智能任务推理时带宽需求高与实时性差、端侧设备计算能力与能量受限的问题,在端侧设备处引入无线射频供能技术以实现端侧设备独立供能,进而提出了基于无线射频供能的端边协同智能任务推理机制,以最大化端侧设备智能任务推理完成率。构建端侧设备能量收集和端边协同推理模型。考虑端侧设备射频(Radio Frequency,RF)能量收集时间与端边协同智能任务推理时间约束、端侧设备传输功率与可使用能量约束及边缘侧计算资源约束,最大化智能推理任务完成数量。提出了一种基于深度确定性策略梯度(Deep Deterministic Policy Gradient,DDPG)的深度神经网络(Deep Neural Network,DNN)模型分割与通信计算资源联合优化算法,以得到最优的端侧设备射频能量采集时间、端侧设备传输功率、DNN模型分割点和边缘计算资源分配。仿真结果表明,所提算法可有效提高智能任务推理完成率,且明显优于其他对比算法。

To solve challenges of high bandwidth demand and poor real-time performance in cloud-based intelligent task inference,as well as limited computing power and energy constraints of edge devices,we introduce a wireless Radio Frequency(RF)energy harvesting technology at the edge devices to achieve independent energy supply.An end-edge collaborative intelligent task inference mechanism based on wireless RF energy harvesting is proposed to maximize the completion rate of intelligent task inference at devices.Firstly,an energy harvesting model and edge collaborative inference model are constructed.Secondly,considering time constraints for RF energy harvesting at edge devices,time constraints for device-edge collaborative intelligent task inference,power constraints for transmission at devices,constraints on available energy,and constraint of edge computing resources,we maximize the number of completed intelligent inference tasks.Finally,a joint optimization algorithm based on Deep Deterministic Policy Gradient(DDPG)is proposed to optimize partition decision of Deep Neural Network(DNN)models and allocation of communication and computing resources at edge,aiming to obtain optimal RF energy harvesting time,transmission power,DNN model partition points,and edge computing resource allocation.Simulation results demonstrate the effectiveness of the proposed algorithm in improving the completion rate of intelligent task inference,which outperformes other comparative algorithms significantly.