Research On Parameter Identification And Attitude Control Of Combined Spacecraft Based On Deep Learning

The capture of non-cooperative targets exists in many scenarios of on-orbit servicing and space confrontation.The attitude control of the combined spacecraft after capture will face great difficulties because the non-cooperative target is out of control or has no willingness to cooperate,its dynamic characteristics are unknown,and has residual relative angular velocity and linear velocity,In this paper,the attitude control system of the combined spacecraft is been taken as research object to deeply study the parameter identification and control of the combined spacecraft.Firstly,an intelligent parameter identification algorithm based on convolution neural network is studied,to solve problems such as unknown mass,centroid position and moment of inertia parameters of the combined spacecraft by taking advantages of deep learning in multi-parameter optimization.The algorithm could identify multi-parameters of combined spacecraft in condition of the no conservation of linear momentum and angular momentum,having more general adaptability than that of zero external force.A 4-layer convolution neural network are designed by using the character of the weight sharing of convolution neural network.The identification of inertial parameters with high precision was achieved by plenty training of state data in a specific form of storage in a short time.The simulation results show that the parameter of the mass,centroid position and inertia matrix of combined spacecraft could be accurately and quickly identified under the interference of external random force and moment:the mass and the position of the center of mass is identified in 24s,the rotational inertia parameter is identified in 1190s,the identification accuracy is within 3%.Secondly,a model predictive control algorithm based on convolution neural network with weak computational power is studied to reconstruct the attitude control law of combined spacecraft in multi-scene.Considering the weakness of the spacecraft's hardware processing ability,the algorithm first uses model predictive algorithm to control the combination from the initial state to desired position,and then using the state variables to train the three-layers convolutional neural network.Convolutional neural network can replace model prediction control to control the combination after trained.The simulation results show that the algorithm can predict the control parameters within five control cycles,hardware calculation time of the algorithm is reduced by about 5.5 times compared with the traditional model prediction algorithm.Attitude control of the combination can be completed in 30 seconds;the control accuracy is 10^-4 orders of magnitude.Finally,the validity of the identification algorithm and model prediction algorithm based on deep learning in the integrated control system is verified by the analysis of actual attitude data from the distributed digital simulation platform..The control algorithm researched in this paper combines the deep learning technology with the traditional control algorithm,taking advantage of the advantages of deep learning in large data processing,and having more effective control ability in multi-task scenarios.It provides an idea for the application of deep learning in spacecraft control system in the future.