Application of GA-based neural network for attitude control of a satellite

Republic of China Satellite (ROCSAT-1) is the Republic of China's first low Earth orbit scientific satellite. The motion and stability of the satellite can be influenced by internal and external interferences. The neural network can learn the characteristics of systems according to the training samples taken from the I/O sampling data of the systems without complex mathematic models. We screened the suitable training samples by eliminating poorer data and adding better data according to the input–output relationship to obtain good performance of the controllers. Genetic algorithms were used to find the optimal initial value of neural network parameters. To show the controller's performance in restraining disturbance and maintaining stability, simulation was carried out under an interference scenario.