Intention Recognition of Maneuvering Spacecraft Cluster Using MIC-Net

Spacecraft clusters play a crucial role in space missions. Accurately identifying their motion intentions is essential for reducing collision risks and enhancing space traffic management. However, current research on spacecraft cluster intention recognition remains limited. To address this gap, we define 11 representative motion intentions for spacecraft clusters and construct eight maneuvering scenarios through the application of impulse maneuvers. Intention recognition under maneuvering conditions poses additional challenges due to frequent state changes, varying intentions, and increased complexity. To tackle these issues, we propose a maneuvering intention and cluster analysis network (MIC-Net). This three-layer network architecture effectively decomposes the problem and enables reliable cluster intention recognition. The MIC-Net contains three layers, i.e., maneuver layer, intention layer, and cluster layer, which, respectively, use the interacting multiple model with labeled multi-Bernoulli (IMM-LMB) algorithm and the bidirectional gated recurrent unit multihead attention (BiGRU-Multi-Head Attention) neural network. The IMM-LMB algorithm enables the detection of maneuvers in multiple targets while maintaining robust and efficient state estimation. The BiGRU-Multi-Head Attention neural network model is capable of classifying time-series data and demonstrates strong generalization ability. This approach is then applied to perform intention recognition and cluster identification. In the simulation section, the reliability and robustness of MIC-Net are verified through eight maneuvering scenarios of spacecraft cluster. Meanwhile, the results also demonstrate the superiority of BiGRU-Multi-Head Attention in terms of both accuracy and loss function compared with other methods for single-target intention recognition. The proposed neural network is further applied to cluster identification tasks, achieving high classification accuracy.