Conjunction Screening Method for Cataloged Space Objects Based on GCN Model

As the space environment becomes increasingly congested, efficiently identifying and mitigating potential on-orbit collisions has become critical for Space Traffic Management (STM). However, the rapid growth in the number of Resident Space Objects (RSOs) presents significant challenges for accurate collision risk assessment. Advanced conjunction screening methods are essential to enhance the efficiency of collision analysis. In this study, we formulate the all-vs-all conjunction screening as a link prediction problem within the framework of Graph Neural Network (GNN). Specifically, we develop a Graph Convolutional Network (GCN) model to capture graph topology features, thereby identifying spatial dependencies between nodes. The exper-imental findings indicate that the proposed GCN model attains an F1-score of 0.8847, successfully forecasting con-junction events involving RSOs and outperforming current approaches.