Mach number prediction models based on Ensemble Neural Networks for wind tunnel testing

The 2.4m×2.4m wind tunnel is a system with the properties of strong nonlinear, multiple variables, serious coupling, large lagging, time-varying, etc. The complexity of all these phenomena makes the development of suitable dynamic Mach number models based on the aerodynamics laws very difficult. As an alternative, the Ensemble Neural Networks (ENN) model based on the feature subsets is proposed to address this problem. ENN built the sub-models on different lower dimension data sets, and reduced the complexity of the single Neural Networks (NN) built on the whole data set. Furthermore, a comparative study among the single NN models and the ENN models when used to predict the Mach number is conducted. Results show that the performance is improved by the ENN models. It is also shows that training time and testing time are much reduced by the ENN models.