Twin Support Vector Regression and Its Application on Aerodynamic Design

Abstract

AbstractSurrogate modeling is playing an increasingly important role in multidisciplinary design optimization (MDO) related to different areas of aerospace science and engineering, Support vector regression (SVR), due to its good behavior related to numerical noise filtering and highly nonlinear function modeling, is promising as an alternative modeling method. However, SVR is time-consuming for high dimension large-scale samples problem. Since Twin support vector regression (TSVR) method shows faster modeling speed, this work aims to evaluate the modeling abilities by numerical examples, and introduce the TSVR method into aerodynamic designs to explore its potential in the aerospace science. Through series of numerical examples, including low-dimensional numerical examples, high-dimensional numerical examples and numerical examples with noises, it is shown that TSVR takes much less time for modeling while keeping high modeling accuracy. Then, taking RAE2822 airfoil as an example, TSVR and SVR is compared, in which TSVR still shows higher modeling efficiency and accuracy. It’s preliminarily proven of the TSVR has great potential in aerodynamic design.KeywordTSVRSVRSurrogate modellingMDO