A seq2seq model based on autocorrelation attention for long-term orbit prediction using two-line element

Accurate long-term orbit prediction of space objects is essential for mission planning, collision avoidance, and ensuring mission safety and success. Current models often fail to fully account for external forces like atmospheric drag and radiation pressure, leading to a significant degradation in prediction accuracy over time. To address this issue, this paper proposes a seq2seq model based on autocorrelation attention (SMAA) designed for long-term orbit prediction. The model first encodes input sequences and then applies an auto-correlation attention decoder that emphasizes key information context to improve orbit prediction. Experimental results show that the SMAA achieves a 27.93 % reduction in mean square error compared to the Support Vector Machine and Long Short-Term Memory networks. Furthermore, The SMAA method demonstrates flexibility in orbit prediction tasks, showing adaptability to various orbital regimes and data conditions. This study offers a novel approach to orbit prediction, providing technical support for the future space mission planning and the maintenance of space safety. (c) 2025 COSPAR. Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.