Application of RBF neural network ensemble to aerodynamic optimization

To reduce the expensive computational cost in optimization, surrogate model is suggested to evaluate objective function. However, for its prediction error surrogate model is likely to mislead the search in optimization. To avoid this danger, on one hand, improving prediction ability of surrogate model is necessary. In this paper, neural network ensemble method is used to reduce prediction error of single RBF neural network. On the other hand, effective management of optimization algorithm and surrogate model is needed. In this paper, MADS optimization algorithm and RBF neural network ensemble are integrated effectively. The test cases show that this method is efficient and effective for aerodynamic shape optimization.