Objective In modern naval warfare, trajectory prediction and intent recognition of adversarial unmanned surface vehicle (USV) swarm can provide critical decision-making support for combat command systems, thereby enabling strategic advantages in adversarial engagements.

Method This paper proposes a predictive model based on a stacked temporal graph convolutional network (STGCN) and a spatio-temporal Transformer network (abbreviated as GST-Transformer) to forecast future trajectories and recognize intents of adversarial USV clusters. In the GST-Transformer framework, the STGCN is designed to capture temporally correlated interaction features from the historical trajectories of USV clusters, and the spatio-temporal Transformer employs a dual-channel encoder mechanism to extract and fuse spatio-temporal features of adversarial trajectories in parallel. Subsequently, a generative trajectory encoder synthesizes trajectory prediction representations by integrating interaction features and spatio-temporal characteristics. Finally, a multi-head decoder generates both the predicted future trajectories and intent recognition results for adversarial USV clusters.

Results Experiments were conducted using simulated adversarial data of USV clusters. The results indicate that the proposed model achieves superior prediction accuracy in trajectory forecasting tasks, with average displacement error (ADE) reduced by 25.72% and final displacement error (FDE) reduced by 16.27% compared to mainstream methods such as Informer and GRU. Additionally, the model demonstrates high accuracy and discriminative capability in combat intent recognition tasks.

Conclusion The proposed GST-Transformer model exhibits efficient trajectory prediction and intent recognition capabilities for USV clusters, offering novel technological support for modern naval warfare command systems. This advancement enhances situational awareness and decision-making precision in dynamic combat environments.