无人机空地网络算力资源预测与分配联合算法研究

A Unified Approach to Computing Resource Prediction and Allocation in UAV Air-Ground Networks

无人机空地网络相较于地基网络可以显著提高网络覆盖率。空地网络动态性强、拓扑变化复杂,传统的算力资源分配算法已无法满足网络需求。针对上述问题,提出了一种基于数字孪生技术和多智能体强化学习的算力需求预测与动态资源分配联合算法,首先建立无人机空地网络模型,之后搭建其数字孪生网络并基于长短时记忆网络预测网络算力需求,最后基于多智能体强化学习算法实现算力资源的动态分配。仿真结果表明,所提算法可以有效改善网络性能。

Compared to ground-based networks,unmanned aerial vehicle(UAV)-ground networks can significantly enhance network coverage.However,the dynamic nature and complex topology of air-ground networks present challenges that traditional computing resource allocation algorithms cannot adequately address.To tackle these issues,we propose a joint algorithm for computing resource demand prediction and dynamic allocation,leveraging digital twin technology and multi-agent reinforcement learning(MARL).Initially,a UAV-ground network model is established,followed by the construction of its digital twin.The computing resource demand is then predicted using a long short-term memory(LSTM)network,and dynamic resource allocation is achieved through a MARL algorithm.Simulation results demonstrate that the proposed algorithm effectively improves network performance.