Multi-model fusion surrogates ensemble using promising parameter domain for flight vehicle system design

Multi-model fusion surrogate has caught significant attention in flight vehicle system design, owing to its ability to tradeoff the approximation accuracy and computational efficiency in recent years. To further improve the performance of multi-model fusion surrogate, an ensemble of multi-model fusion radial basis function surrogates using promising parameter domain (MFRBF-PPD) is proposed in this paper. In MFRBF-PPD, a series of basis multi-model fusion surrogates are ensembled to improve the approximation accuracy. A concept of promising parameter domain (PPD) is proposed to select proper parameters. And the basis multi-model fusion surrogates are constructed according to the hyperparameters in PPD. A synthesized discrepancy index is defined to efficiently calculate the weight coefficient for each basis multi-model fusion surrogate. Numerical benchmarks test results indicate that the proposed MFRBF-PPD performs better on approximation accuracy and robustness than the traditional single-fidelity surrogates and competitive multi-model fusion surrogates. Finally, MFRBF-PPD is applied to a ball head blunt cone heat flux prediction problem and a solid rocket motor multidisciplinary design optimization problem. Results illustrate the effectiveness and practicability of MFRBF-PPD for real-world engineering applications.