Orbit Prediction And Collision Probability Calculation Method Of Space Station Based On Intelligent Algorithm

The construction of space station is an advanced stage in the development of manned spaceflight.Due to the large size of the space station and its general operation in low Earth orbit,its motion is greatly affected by the drag perturbation of the upper atmosphere,and the probability of collision with space debris is also high.The density of the upper atmosphere is affected by a variety of complex factors and cannot be accurately modeled at present,which affects the orbit prediction of the space station and the calculation of the probability of collision with debris.In this dissertation,classical orbital dynamics theory is combined with new artificial intelligence algorithms such as deep neural network and reinforcement learning algorithm to solve the orbital prediction and collision probability calculation problems of space station.Firstly,a method to improve the accuracy of orbit prediction by using DDPG reinforcement learning algorithm to compensate the numerical integration results of space station motion equation is studied.Based on the analysis of the perturbation force of the space station,the equation of its in-orbit motion is established,and the compensation scheme of numerical prediction results based on reinforcement learning is designed.The reinforcement learning process modeling and algorithm selection are completed,the appropriate input and output parameters of reinforcement learning network are determined,and the training and test samples of the network model are generated.The deep learning and DDPG reinforcement learning networks are trained and tested,and the orbit prediction accuracy and improvement effect are analyzed based on numerical simulation.Secondly,PPO2 reinforcement learning algorithm is used to modify atmospheric model parameters,so as to improve the prediction accuracy of numerical integration of dynamics equations.PPO2 reinforcement learning network is introduced on the basis of US1976 standard atmospheric model to compensate the forecast error caused by inaccurate atmospheric model.An orbital prediction scheme based on reinforcement learning is designed to compensate the deviation of atmospheric model parameters.The reinforcement learning network model is designed,the training parameters are selected and the network training is completed.The real-time performance and accuracy of the reinforcement learning method are verified by numerical simulation.Finally,a method to calculate the collision probability between space station and space debris using deep neural network instead of multiplenumerical integration is studied.The encountered scenario between space station and space debris is constructed,the traditional collision probability calculation method based on multiplenumericalintegration is discussed,and a collision probability calculation scheme based on deep neural network is designed.Multiplenumerical integration method is used to generate training and test samples of network model,and the training of deep neural network model is completed.The effectiveness and rapidity of the proposed artificial intelligence method are verified by joint numerical simulation of orbit prediction and collision probability calculation.The research work of this paper has reference value for the research and engineering implementation of orbit prediction and safe operation of China’s space station and low orbit spacecraft.