XGBOOST APPLIED TO MODEL AN AIRCRAFT ENGINE RPM-FUEL RELATIONSHIP THROUGH NONLINEAR BLAC-BOX SYSTEM IDENTIFICATION

Abstract

In the aerospace industry, it is important to understand the behavior of turbojet engines, which is considered a complex system and, in many cases, engine’s performance data is not available, causing lack of knowledge about the system performance. In order to meet this requirement, black-box system identification strategies are been successfully applied complex engineering problems. Based on the previous assumptions, this paper presents a parametric black-box system identification approach based on data collected from a turbojet engine through an experimental investigation. In order to model the jet engine, Extreme Gradient Boosting (XGBoost) technique, which combines a series of regression trees, was associated to a nonlinear autoregressive exogenous (NARX) model structure. In order to evaluate the number of regressors associated to the model, the relative importance of each parameter was evaluated. Finally, the model was validated considering the coefficient of determination (