Ensemble fixed-size LS-SVMs applied for the Mach number prediction in transonic wind tunnel

It is important to predict the Mach number in a transonic wind tunnel system. According to the aerodynamic mechanism, the Mach number is indirectly calculated from the total pressure (TP) in the stilling chamber and the static pressure (SP) in the test section. The high-dimensional input features and large-scale data are the main difficulties to build the TP and the SP models. The fixed-size LS-SVM is a popular method to build a nonlinear model for a large-scale problem. However, it is difficult to further improve the sparsity in the high-dimensional input space. Based on the multivariate fuzzy Taylor theorem, the feature subsets ensemble (FSE) method is proposed to deal with the high-dimensional problem. The set of direct, exhaustive, independent feature-space subdivisions forms the basis to develop FSEs. In the FSE, submodels are learned using substantially low-dimensional data sets and characterized by low complexity. The TP and the SP are estimated with the FSE-based ensemble fixed-size least squares support vector machines (LS-SVMs). Experiments show that the FSEs speed up both training and testing time that would otherwise be infeasible for individual, bagging, and random subspace (RS). FSE models meet the requirements of the forecasting speed, the accuracy and the generalization of the Mach number prediction.