集成神经网络在2.4m跨声速风洞马赫数预测中的应用

Application of Ensemble Neural Network in the Prediction of Maher Nnumber in 2.4 m Transonic Wind Tunnel

2.4 m跨声速风洞空气流动是复杂的三维流动,想要利用机理模型来描述马赫数的特性十分困难,所以采用数据驱动的方式建立风洞马赫数模型。提出一种基于特征子集的集成神经网络建模方法,该方法选用动态NARMAX模型,并采用集成神经网络的方法建立了风洞马赫数预测模型;最后,进行了单一神经网络模型与集成神经网络模型在马赫数预测上的性能对比。试验结果表明:集成神经网络模型可以在保证预测准确度和泛化性的基础上,降低模型的训练和测试时间。

Air flow is a complex three-dimensional flow in the 2.4 m transonic wind tunnel, that is very difficult to use of mechanism modeling to describe the characteristics of Mach number, so we adopt the way of data driven to set up the wind tunnel Mach number model. This paper proposes the ensemble neural networks （ENN） modeling method based on feature subset, the method chooses dynamic NARMAX model, and integrated ENN method to establish a predictive model wind tunnel Mach number. Furthermore, a comparative study among the single neural networks （NN） models and the ENN models when used to predict the Mach number is conducted. Results confirm that training time and testing time are much reduced by the ensemble neural networks models.