Intention recognition for spacecraft formation based on two-layer temporal convolutional network-self attention

Intention recognition for non-cooperative targets is a vital component of space situational awareness. Aiming to address the spacecraft formation intention recognition problem with sunlight constraints, we propose a two-layer Temporal Convolutional Network-Self Attention-based (TTCN-SA) method. First, based on the orbital relative dynamics model, a typical motion intentions set is established. Second, considering the incomplete information caused by sunlight constraints, we use a prediction network to fit the missing motion states. And then, recognition network is employed to recognize the intention with the complete motion information. Finally, we conduct simulation experiments, and the results show the TTCN-SA model has an accuracy of up to 98.37% for formation intention recognition. The recognition accuracy and efficiency are both higher than existing intention recognition methods.