Nonlinear Spacecraft Attitude Control via Cascade-Forward Neural Networks

Abstract

The problem of spacecraft attitude control has been addressed in the literature via several linear and nonlinear algorithms. In the current study, cascade-forward neural networks have been used to mimic the behaviour of the High Performance Nonlinear Discrete Controller (HPNDC). Controller stability is proven via Lyapunov second method. The developed control algorithm has the ability to work during all of the spacecraft operational modes and alleviates many problems associated with other nonlinear algorithms existing in the literature. Performance of the algorithm has been tested against various nonlinear attitude control algorithms including HPNDC, Sliding Mode (SM) Controller, Nonlinear Dynamic Inversion (NDI)controller, and Adaptive Neuro-Fuzzy Inference Systems (ANFIS) controller. The proposed neural networks control algorithm achieved a good performance compared to the aforementioned control algorithms. The behaviour of the HPNDC control algorithm has been used for training of the Neural Networks due to its numerous advantages when operating in linear and nonlinear operational modes. To prevent data overfitting, 10% of the training data set are used for testing. The training ratio is selected to be 90%. Gravity gradient torques acting on EGYPTSAT-1 spacecraft are considered the main acting disturbance. The developed control algorithm can nullify the initial high angular velocities and attitude angles of the spacecraft within two orbits.