Abstract:

In response to the scarcity of available spectrum resources in UAV swarms and the difficulties in solving multi-objective optimization problems,as well as the challenges of obtaining complete channel information and poor real-time performance during the resource optimization process,a knowledge graph-assisted spectrum resource optimization algorithm for UAV swarms is proposed.Firstly,a relation-aware graph multi-head attention network(RGMAN) encoder is constructed to aggregate communication parameters,performance parameters,and electromagnetic environment information of the UAV swarm,and allocate different weights to neighbor information based on the importance of the nodes.Then,an improved layer-attention-based InteractE(SE-IE) model is developed to predict the channel access and transmit power for the UAVs,which utilizes a squeeze-and-excitation module to obtain layer attention information and extracts deep-level interactive information from the results of circular convolutions.The simulation results indicate that the proposed algorithm exhibits rapid convergence capability,excellent performance in link prediction,and notable stability and robustness on public datasets.Additionally,on the dataset for UAV swarm spectrum management,the proposed algorithm can generate an approximately optimal spectrum resource optimization scheme for UAV swarms,in the premise of channel distribution information and partial environmental information.

Key words: unmanned aerial vehicle swarm, resource allocation, knowledge graph, graph neural networks