A modeling method for aero-engine by combining stochastic gradient descent with support vector regression

Abstract

Aero-engine aerodynamic model is widely applied to identify the aerodynamic parameters of components like compressor pressure, turbine temperature and so on. A data-driven modeling method for the aero-engine aerodynamic model by combining stochastic gradient descent with support vector regression (SGDSVR) is proposed. A novel support vector regression (SVR) training mechanism that combines batch learning with online learning is presented according to the demand and characteristic of the aero-engine aerodynamic model. In the training mechanism, batch learning is to build the initial model and online learning is to modify the online model based on the initial model. An improved sequential minimal optimization (SMO) algorithm is introduced during building the initial model phase and the SGDSVR algorithm is proposed during modifying online model phase. The simulation data of an aero-engine component-level model and the flight data of a certain aircraft are used to test the modeling method and the proposed method shows better performance compared with traditional methods.