Performance of Gradient Boosted Trees for Prediction of Coaxial Inflow Models with XGBoost
This paper explores the capabilities of regression modeling with gradient boosted trees in XGBoostTM as a potential solution to counter limitations of real-time analysis in rotorcraft.
Abstract
Finite-state methods are essential for developing accurate and efficient inflow models for flight dynamics in rotorcraft. Recent updates in the field allow for the application of finite-state inflow models to multi-rotor systems using the adjoint theorem, which involves time delays and adjoint variables. However, the addition of time delays and adjoint variables comes with higher computing requirements, limiting the ability of real-time analysis. This paper explores the capabilities of regression modeling with gradient boosted trees in XGBoostTM as a potential solution to counter limitations. The investigation entails data density analysis and XGBoostTM hyperparameter searches to determine the best model and amount of data needed to reach respectable performance for velocity distributions across the lower rotor disk.