Spacecraft attitude control method based on deep neural network approximation MPC

Abstract

The invention discloses a spacecraft attitude control method based on deep neural network approximation MPC, and the method comprises the steps: S01, configuring an MPC controller, controlling a plurality of input attitude parameters to be input into the MPC controller, obtaining a plurality of corresponding control torque outputs, and constructing a training data set through the combination of each input and output; s02, constructing a DNN model, training the DNN model by using the training data set so as to enable the DNN model to approach the MPC controller, and obtaining a target DNN model approaching the MPC controller after the training is completed; and S03, performing attitude control on the target spacecraft by using the target DNN model obtained after training, acquiring real-time attitude parameters of the spacecraft in the control process, inputting the real-time attitude parameters into the target DNN model, and providing the output expected control torque for the spacecraft. The method has the advantages of being simple in implementation method, good in control performance, high in control efficiency, low in calculation complexity and the like.

Description

technical field [0001] The invention relates to the technical field of spacecraft control, in particular to a spacecraft attitude control method based on deep neural network approximation to MPC. Background technique [0002] Model Predictive Control (MPC, Model Predictive Control) has been widely used in industrial practice. By repeatedly solving optimization problems within a given prediction range, it has the advantages of explicitly considering constraints and realizing the optimization of the objective function. . When it comes to constrained control problems, that is, in the case of attitude control, MPC has outstanding performance. However, MPC has high requirements for online computing capabilities, and the performance of on-board chips is very limited due to power consumption. As a result, real-time calculation of MPC on the star often takes too long, that is, it is difficult to directly implement attitude control based on MPC. Especially for low-power on-board co...

AI Technical Summary

Problems solved by technology

However, the structure and scale of the conventional neural network are not suitable for the attitude control of the spacecraft. It is difficult to determine the scale of the neural network in practical applications, and if the DNN is too large, the training is usually difficult and time-consuming. Neural network implementation of spacecraft attitude control still has the problem of time-consuming, and it still cannot meet the efficiency and real-time requirements of spacecraft attitude control.

Images