Cluster spacecraft intent recognition under multi-mode maneuvers

During actual space operation missions, cluster spacecraft may achieve mission objectives through various combinations of orbital maneuvering modes. With the numerous targets in cluster, large maneuvering uncertainty, and rapid changing cluster configurations, there are demanding requirements for the response speed and accuracy of the intention recognition methods. To address this concern, this paper proposed a robust intension recognition method based on the random finite set theory and deep learning methods. Firstly, the identities of multiple cluster spacecraft are managed through the Labeled Multi-Bernoulli (LMB) filter, where the orbital state and the maneuvering parameters are jointly estimated by incorporating the LMB filter with the Interactive Multiple-Model (IMM) approach. The IMM-LMB scheme is able to quickly and accurately identify the maneuvering mode of multiple spacecraft. Subsequently, the trajectory information of maneuvering spacecraft corresponding to different maneuvering modes can be extracted as necessary evidence for decision-making. Secondly, a robust intension recognition method based on Conv-BiGRU is designed by integrating spatiotemporal features. Through the extraction of spatial and temporal correlation features from multivariate data, the reconfiguration of clustered spacecraft caused by maneuver can be evaluated and the intentions of clustered spacecraft are determined. In order to verify the developed method, a simulated case study is developed, where various intentions, e.g., encirclement, escape, on-orbit servicing and dispersion camouflage, of cluster spacecraft are generated based on CW equation. The designed IMM-LMB method can quickly and accurately identify the maneuvering situation of targets, and the developed neural network can identify the intentions of the clustered spacecraft with an accuracy of more than 93%.