Redka M. Models and methods of machine learning for dynamics analysis and control of spacecraft during space debris removal using ion beam

At this time, a large number of objects that do not perform useful functions and are called space debris have accumulated in Earth orbits. According to scientists, further human activity is impossible without the introduction of methods of active space debris removal. One of the proposed methods is the method of removal using an ion beam, which transmits a force impulse to the space debris object to remove it from orbit. However, it is known that the implementation of such removal systems is complicated due to the high cost and development time. At the same time, hardware and software methods of machine learning, including deep learning and reinforcement learning, have been developed and have proven themselves in tasks related to control, navigation, and image analysis. The study is devoted to improving the efficiency of key tasks that need to be solved for the application of the ion beam space debris removal method using modern models and machine learning methods. The proposed methods are compared to traditional ones in terms of accuracy and speed. For the task of determining the force of the ion beam impact, it is determined that the proposed methods allow determining the force at least 2 times faster than the traditional (analytical) method using central projections on an auxiliary plane. The equations for the computational complexity of neural network algorithms in comparison with the analytical one are obtained. For the task of controlling the orientation of the spacecraft, the methodology of reinforcement learning is applied to obtain a control law, which can be improved as the spacecraft interacts with the environment. For the task of controlling the relative position of the spacecraft, a control law was obtained taking into account engine activation, which allows reducing fuel consumption by the propulsion system. The obtained reinforcement learning control methods are compared with the traditional ones.