Artificial Neural Network Based Initial Orbit Determination for Angle-Only Data

The initial orbit determination methods are well-established for various kinds of observational data including angular, range, and range-rate data. In this work, we investigate the use of artificial neural networks (ANNs) for the initial orbit determination with angle-only observations. We used three regression algorithms of MATLAB to train ANNs predicting semi-major axis and eccentricity of equatorial orbits. The trained ANNs can predict semi-major axis values with a mean error of 0.0021% and eccentricity values with a mean error of 0.0152%. These results are obtained for 200 angular data for a given observation pass. The real promise for ANN based approach is that the raw observational data can possibly be used without any preprocessing. The results obtained in this study provide initial steps in that direction.