Fuzzy neural control of satellite attitude by TD based reinforcement learning

With recent development of the space science and technology, higher requirements such as accuracy, robustness and disturbance rejection ability in satellite attitude control system have leaded to the more promising intelligent control methods. In this paper, a fuzzy neural control approach applied to the three-axis stabilized satellite is presented. In order to solve the problems of online learning and tuning of the fuzzy neural network parameters, the reinforcement learning based on temporal difference (TD) is also proposed and studied so that the training samples for the self-learning controller are not needed. Since the vibration of the solar swing cannot be ignored, a flexible mathematic model of the satellite is studied, employing Quaternion and Euler-Angles representations. The simulation results showed that the proposed control method with reinforcement learning architecture could not only improve the accuracy and robustness of the system, but also could deal with the uncertainties and external disturbance efficiently.