基于证据理论的多航天器轨道博弈意图识别方法

For the multi-spacecraft orbital game intention recognition problem, a three-layer inference method of “parameter prediction-game behavior-game intention” is proposed by integrating deep learning and evidence theory. First, the correspondence between game behaviors and intentions of the target spacecraft cluster is established based on expert experience. Second, a deep neural network with Transformer modules is designed to extract time-series orbital state features and predict key behavior parameters. Then, evidence theory is applied to map behavior parameters to spacecraft game behaviors. Finally, an intention inference strategy based on conditional probability tables is developed to deduce cluster intentions from game behaviors. Comparative experiments with end-to-end deep learning models demonstrate that the proposed method achieves high recognition accuracy while significantly improving the interpretability of the intention recognition process, providing theoretical support for intelligent decision-making in complex orbital game scenarios.