| Popis: |
Aerodynamic heating is a critical problem to consider in hypersonic flight. It involves many factors, and most of them affect the result nonlinearly, which makes it difficult to get a proper model from experimental data. Even worse, it is hard to gather enough data for distilling a model since aerodynamic heating experiments are costly. Machine learning (ML) methods are possible candidates for its data modeling. However, general ML needs more data for modeling. Therefore, a ML strategy that can capture strong nonlinear relations with small-size dataset is desirable. In this work, a special ML strategy that aims at modeling data collected from hypersonic aerodynamic heating experiments is established. The strategy is based on the randomized neural network (RNN) whose basic model framework is a single-hidden layer feedforward neural network (SLFN). A global optimization (GO) technique, low dimensional simplex evolution (LDSE), is introduced to improve its correlation performance. The modified algorithm is referred to as LDSE enhanced RNN for short, in which the weights and biases in the hidden layer are globally optimized, rather than randomly generated. Theoretically, the LDSE enhanced RNN has the hierarchically global optimality. Meanwhile, the LDSE enhanced RNN has been applied to model a real word aerodynamic heating database of blunt-body. Study shows that the LDSE enhanced RNN has a good capability to balance the complexity and accuracy of a nonlinear regression model, and the model can give a reliable estimation of the aerodynamic heating. |
