HMAPPO:无线供能边缘计算网络长时吞吐量最大化方法OA

With the continuous deployment of Internet of Things(IoT)and 5G networks,the computational load and sustainable energy demand of edge sensor devices have increased significantly.By integrating Wireless Power Transfer(WPT)and Mobile Edge Computing(MEC),Wireless-Powered MEC(WP-MEC)provides a promising solution for extending the power supply lifetime of edge devices and enhancing overall system computing capability.However,previous works focus on single-time-slot resource optimization or single-cell network models,leading to low resource utilization efficiency and significant deviations from practical scenarios.To address this issue,the optimization of a multi-cell and multi-time-slot WP-MEC network based on Non-Orthogonal Multiple Access(NOMA)is focused by jointly optimizing energy transmission time,task offloading strategies,and power allocation,and energy accumulation gain to maximize the long-term system throughput is fully leveraged.To enable efficient resource scheduling in complex and dynamic networks,a Heterogeneous Multi-Agent Proximal Policy Optimization(HMAPPO)algorithm is proposed.By introducing a hierarchical structure with a controller agent and device agents,HMAPPO achieves cooperative optimization between global energy transfer time,local task offloading,and power allocation.Unlike value-function-based approaches such as Multi-Agent Soft Actor-Critic(MASAC)or Multi-Agent Twin Delayed Deep Deterministic Policy Gradient(MATD3),HMAPPO adopts a proximal policy optimization mechanism that constrains changes between successive policies.This makes it more suitable for multi-slot energy dynamics and continuous action spaces,thereby achieving higher training stability in WP-MEC networks.Simulation results demonstrate that the proposed algorithm achieves performance comparable to that of the centralized Proximal Policy Optimization(PPO)while realizing distributed optimization,with a performance gap of less than 3.3%.Moreover,the algorithm exhibits performance under varying conditions of different numbers of cells,devices,and device distances,verifying its superior generalization and scalability.