Regularized Regression Techniques for Model Reduction in Spacecraft Thermal Analysis

This study addresses two primary challenges in spacecraft thermal analysis: the need for onboard computation to support autonomous operations, and the requirement for real-time processing capabilities that allow thermal analysis to be seamlessly integrated into simulators for mission planning. It focuses on overcoming computational limitations on low-capacity onboard computers, traditionally reliant on model reduction methods suited for ground-based applications. The paper reassesses these methods, distinguishing between model-based and data-driven strategies, with a core focus on discrete thermal equilibrium equations formulated as linear combinations of inputs and temperature differentials. It introduces a novel model construction technique using Lasso regression for data preprocessing and feature engineering within thermal analysis contexts. This is validated on a simplistic four-node thermal model and a 6U-sized satellite from the University of Tokyo. The approach demonstrates precise long-term temperature predictions in nonequilibrium states. The post-training analysis of the regression model highlights its high interpretability and reliability, two crucial factors for the success of space missions. The method’s adaptability for onboard implementation, particularly on microcomputers, underscores its efficacy in achieving high prediction accuracy and computational time reduction, thus advancing space thermal analysis and enhancing spacecraft operational safety and efficacy.