Research on LEO Satellite Orbit Prediction Based on Patch-Transformer

Low earth orbit satellites have limited accuracy in medium and long-term orbit prediction owing to uncertainties in atmospheric drag modeling. To address this limitation, an intelligent orbit prediction method based on the pseudo-drag coefficient and Patch-Transformer neural network is proposed. The pseudo-drag coefficient transforms the atmospheric drag modeling problem into a time-series prediction task. The Patch-Transformer neural network captures the complex mapping relationship between the pseudo-drag coefficient and space environment parameters, thereby establishing an orbit prediction framework that integrates physical models with data-driven approaches. The experimental results—based on the Swarm A satellite—demonstrated that, compared with traditional method, the proposed method reduced the median orbit position error by approximately 23% within a 7 d prediction period. Validation under geomagnetic storm and solar flare scenarios further indicated that the proposed method has a certain degree of applicability under complex space environment conditions.