Multidimensional Trajectory Prediction of UAV Swarms Based on Dynamic Graph Neural Network

The advent of AI and 5G technologies has markedly enhanced the intelligence and connectivity of UAVs, leading to the development of UAV swarms. These swarms not only exhibit superior efficiency and adaptability in collective tasks but also offer considerable potential in both civilian and military sectors. However, despite the innovative insights provided by UAV swarm networking in trajectory forecasting, current approaches face obstacles due to the inherent dynamic complexity of these swarms, often neglecting the data from inter-swarm interactions. This research begins by defining metrics of link channel capacity to record the informational exchanges within UAV swarms, thus laying the foundation for a network of UAV swarms. It then advances a dynamic graph neural network (DynGN) model that utilizes an encoder-decoder structure combining a graph convolutional network with a gated recurrent unit. This model processes both the evolving network configuration and trajectory data of UAV swarms simultaneously, enabling more precise trajectory predictions. Through experiments focusing on prediction accuracy, node number stability, and noise robustness, the effectiveness of the model is assessed. Results indicate that the DynGN model outperforms conventional trajectory prediction models, achieving notable improvements in accuracy and fit quality. Moreover, its robustness against noise in dynamic trajectory data highlights its extensive utility in practical mission contexts.