Parameter Identification Of Spacecraft Control System Based On Deep Learning

With the vigorous development of space station technology and Lunar Exploration Project and Mars Exploration Project,space missions have become more complex with longer working hours in orbit and higher requirements for spacecraft control accuracy.The control accuracy of the spacecraft control system is affected by many factors and the parameters of the system are one of the important influencing factors.The parameters of the spacecraft control system will be accurately calibrated on the ground.However,the severe vibration and temperature changes experienced by the spacecraft during launch,in-orbit and in orbit operation may cause the parameters to be different from the results of the ground calibration.In order to solve the problem that the parameters of spacecraft control system are difficult to obtain accurately in orbit,this paper combines cutting-edge deep learning algorithms to identify the spacecraft control system parameters,and mainly studies the measurement error calibration of the attitude determination sensor gyro,the identification of the spacecraft moment of inertia and parameter tuning of PID controller.Main tasks are as follows:The measurement error calibration of the attitude determination sensor gyro is studied.In order to improve the accuracy of attitude control,the measurement error of the low-precision sensor gyro is calibrated on-orbit based on the measurement data of the high-precision sensor star sensor.First,establish the gyroscope’s measurement model and error model and derive the relationship between the gyroscope’s measured angular velocity and the true angular velocity based on the error model.Secondly,the convolutional neural network(CNN)is designed to learn the real angular velocity data and the angular velocity data containing the gyroscope measurement error.Based on the characteristics of the weight sharing of the CNN network,the gyroscope measurement error is obtained through the convolution kernel.The simulation data shows that the CNN network can achieve a good calibration effect.The training can be completed within 39 s and the minimum mean square error can reach 0.2%.Among them,the Adam optimization algorithm can effectively update the network parameters and achieve better convergence effects.The identification of the moment of inertia of the spacecraft is studied.First,establish the spacecraft attitude dynamics equation and derive the observation equation for the identification of the spacecraft’s moment of inertia.Secondly,the recursive least square method is used to identify the moment of inertia,while considering the influence of the error in the angular velocity observation data.Third,a four-layer convolutional neural network is designed to identify the moment of inertia.Through the learning of the angular velocity data and the stored data of the control torque,the moment of inertia is identified by using the characteristics of the weight sharing of the convolutional neural network.The parameter tuning of PID controller is studied.First,the traditional PID attitude controller is designed to control the attitude of the spacecraft.Secondly,respectively design BP neural network and RBF neural network to adjust the parameters of PID controller.The results show that the PID controller based on neural network can effectively solve the problem of unsatisfactory control effect caused by the change of control target or the change of system parameters.