Deep Learning-Based Nonlinear Model Predictive Control of the Attitude Manoeuvre of a Barbell Electric Sail Through Voltage Regulation
Xiaolei Ma, Lei Huang, Hao Wen, Shidong Xu
Abstract: An electric solar wind sail (electric sail or E-sail) is a new type of propulsion system that does not require propellant. This paper proposes a nonlinear model predictive control law based on deep learning to control the attitude of a barbell electric sail. A barbell E-sail is composed of two tip satellites connected via long conductive tethers to a central insulated confluence point. The two tethers are insulated from each other at the confluence point such that the voltages of the two tethers can be independently controlled. The attitude dynamics of the system is modelled using a nonsingular description of tether orientation. To reduce online computation loads, the proposed control scheme has a two-stage design, namely, an offline stage and an online stage. In the offline stage, a large amount of data is generated using the nonlinear model predictive control law and then taken as a dataset to train a deep neural network. In the online stage, feedback control of the system attitude is achieved with extremely low computational cost by performing real-time mapping from the system state to the control output using the trained deep neural network. Finally, the efficacy and performance of the proposed deep learning-based control law are demonstrated via numerical case studies.
原文链接：https://www.sciencedirect.com/science/article/pii/S009457652200073X?via%3Dihub
