A Data-Driven Approach to Spacecraft Attitude Control Using Support Vector Regression (SVR)

This study introduces a novel approach to spacecraft attitude control by employing Support Vector Regression (SVR) as the underlying data-driven controller. Given the inherent nonlinearity of spacecraft dynamics, traditional control design approaches encounter significant challenges. To address this issue, we propose the utilization of the offline data-driven SVR method to synthesize a controller tailored to spacecraft attitude dynamics. The SVR controller is developed through supervised learning methodologies, utilizing offline datasets obtained from the input-output dynamics of a closed-loop system governed by a feedback linearization-based controller. Additionally, SVR can be easily and effectively implemented with nonlinear kernel functions, further enhancing its applicability to complex nonlinear systems. Our findings demonstrate that under realistic conditions, while asymptotic stability may not be guaranteed, the SVR controller ensures the boundedness of closed-loop dynamics. Moreover, we propose a step-by-step method to guarantee a specified steady-state error ball radius based on the convexity property of SVR optimization problem. Extensive simulation studies reveal that the SVR controller exhibits robust performance even in the presence of uncertainties in satellite parameters and external disturbances. A comparative analysis with existing machine learning-based controllers, particularly the Light Gradient Boosting Machine (LightGBM) method, and Deep Neural Network (DNN) highlights the advantages of SVR in terms of control effectiveness and adaptability to varying system conditions. Overall, this research contributes to the advancement of data-driven control strategies for spacecraft applications, emphasizing the efficacy of SVR in addressing nonlinear dynamics and improving closed-loop performance.