ARTIFICIAL INTELLIGENCE LOGIC CONTROL FRAMEWORK AND IMPLEMENTATION METHOD OF SATELLITE GNC SUBSYSTEM

In this paper, an artificial intelligence logic control framework of GNC subsystem is proposed. The machine learning method is used to realize satellite autonomous orbit implementation task. In this framework, the satellite is divided into logical control layer, structure selection layer and support algorithm layer. The logic control layer is the main content of this paper. If the target is not specified manually, the target tendency is established and the intelligent state transfer decision is made. The output value of the intelligent state transfer decision is the state that the system should have in the next step, which is the selection of some certain combination of the structure selection layer. The trend prediction module directly predicts the success probability of the final target after a one-step state hypothesis. The tree search method is used to combine the state transfer decision making module and the trend prediction module. The probability statistics of multiple branches are calculated, and the optimal branch is selected to form the final control sequence and output as the result. The algorithms mentioned above mainly use artificial neural network and Softmax regression, while the reinforcement learning being included in the training process. Random forest is used to predict and diagnose non-state key faults. The system functions are demonstrated through two examples.