Replacement of In-Orbit Modern Spacecraft Attitude Determination and Estimation Algorithms with Neural Networks

摘要

In-orbit spacecraft attitude determination or estimation algorithms are considered to be of utmost importance and criticality. In the current research article, several In-Orbit attitude determinations and estimation algorithms are reviewed. These algorithms are TRIAD, Q-method, and Extended Kalman Filter (EKF). Three analogous algorithms are developed based on Cascade-Forward Neural Networks (CFNN) in order to replace these traditional in-orbit estimation and determination algorithms. In order to prove the concept, the former Egyptian spacecraft EGYPTSAT-1 is utilized as a test case. The objective of the developed CFNNs is to mimic the performance of the presented attitude determination and estimation algorithms. The performance of the proposed CFNNs is evaluated. The proposed CFNN algorithms have the same accuracy level of TRIAD, Q-method, and EKF. The maximum error achieved has been ±0.1° with an average execution time about half of the average execution time of EKF. The proposed CFNN are proven to be generic and could replace any determination or estimation algorithm within the same accuracy levels. The computational load of CFNN is independent of the complexity level of the attitude determination or estimation algorithm that it tries to mimic. This enables the usage of more elaborated models to increase the accuracy within the same computational load.