Satellite Orbit Prediction Based on Recurrent Neural Network using Two Line Elements

Because of the hazards and challenges of the space environment, Satellites are

usually exposed to orbit deviation, collisions with debris, or loss of tracking control.

Therefore, orbit prediction can be defined as the critical and significant role for satellite

monitoring and tracking control. This paper proposes a novel orbit prediction approach

based on Two-Line Elements (TLE) using A Recurrent Neural Network (RNN) architecture

with Long Short-Term Memory (LSTM). The proposed approach has been verified and

evaluated its efficiency using the popular benchmark Clark tracks that describe the orbital

satellites datasets. In the experimental study, the results show that the proposed approach

can predict satellite orbits with high accuracy, which is presented by the two variables,

position and velocity. The evaluation measured are R2 represents the goodness of fitness for

the prediction accuracy is 98%, and the mean square error in position is 9.7∗10−5 and in

velocity is 10∗10−3