无人机网络的用户关联与神经网络模型云边协同剪枝算法  

User Association in UAV Networks and Cloud-edge Collaborative Pruning Algorithm Based on Neural Networks

作　　者：杨健 陶金玉 王成 杨清海[2] YANG Jian;TAO Jinyu;WANG Cheng;YANG Qinghai(The 54th Research Institute of CETC,Shijiazhuang 050081,China;School of Telecommunications Engineering,Xidian University,Xi'an 710071,China)

机构地区：[1]中国电子科技集团公司第五十四研究所,河北石家庄050081 [2]西安电子科技大学通信工程学院,陕西西安710071

出　　处：《无线电工程》2025年第3期633-641,共9页Radio Engineering

摘　　要：针对无人机网络和边缘智能领域中无人机存储受限、网络整体性能不足等问题,提出了基于强化学习的用户关联策略和神经网络模型的云边协同剪枝算法。基于强化学习优化无人机网络拓扑结构,提出无人机用户与基站的关联算法。根据用户关联算法的输出划分网络的云端和边缘端,并筹划神经网络模型的分配类型和模型的大小需求,进一步提出云边协同剪枝算法。充分利用无人机个体的本地数据,对无人机网络协同的神经网络模型进行精简,降低了通信消耗,进一步提高了无人机网络整体性能。结果显示,基于强化学习的用户关联策略提高了信道容量。与使用传统的集中式剪枝精简模型相比,云边协同剪枝算法占用更低上行通信链路上的资源。所提方法在无人机协同环境下实现了高效数据传输和计算处理,为解决资源受限下的通算一体化挑战提供了有效途径。In order to solve the problems of limited UAV storage and insufficient overall network performance in the field of UAV network and edge intelligence,a cloud-edge collaborative pruning algorithm based on reinforcement learning user association strategy and neural network model is proposed.An association algorithm between UAV users and base stations is proposed,which optimizes the UAV network topology based on reinforcement learning.According to the output of the user association algorithm,the cloud and edge of the network are divided,and the allocation type of the neural network model and the size requirements of the model are planned.Furthermore,a cloud-edge collaborative pruning algorithm is proposed.By fully utilizing the local data of individual UAV,the neural network model of UAV network collaboration is streamlined,communication consumption is reduced,and the overall performance of the UAV network is further improved.The results show that the user association strategy based on reinforcement learning improves the channel capacity.Compared with the traditional centralized pruning model,the cloud-edge cooperative pruning algorithm consumes lower resources on the uplink communication link.The proposed method achieves efficient data transmission and calculation processing in a UAV collaborative environment,and provides an effective way to solve the challenges of communication and computing integration under resource constraints.

关 键 词：无人机网络 边缘智能 用户关联策略 神经网络模型精简 

分 类 号：TN929.5[电子电信—通信与信息系统]