Research And Application On Sequential Approximate Optimization Of Trust Region Method Based On Interpolating Ensemble Of Surrogate

In recent years, with the engineering design problems becoming more and more complicated, surrogate based optimization has become popular. Surrogate model is also called as metamodel or responding surface, which is constructed in support of restricted sampling information and can replace the physical model of complex problems. For the optimization processes containing black-box problems, surrogate models can be used to replace such problems to participate in the subsequent optimizations, which can decrease computational cost greatly.As there are lots of methods to construct a surrogate, it's difficult to decide which method to choose in practical using. Currently, there are always many surrogates to be constructed first and the good ones will be reserved while the bad ones be discarded, which can cause big waste in computational resources. The appearance of ensemble of surrogates(EOS) can handle the trouble well. EOS is more robust compared with single surrogate. It is the linear weighting of several single surrogates and always exists the disadvantage of not precise enough. To handle this defect, a new method of constructing an EOS calling interpolating ensemble of surrogate(IEOS) is proposed in this paper. Several benchmarks prove that IEOS can not only inherit the advantage of traditional EOS, but also improve the precision remarkably.Also, a new sequential approximate optimization method which combines the IEOS with the trust region method(TRM) is put forward in this paper. The interpolation ensemble of surrogate based trust region method(IEOS-TRM) is proved to be both rapidly in converging speed and more robust through the benchmark.On the base of the above-mentioned researches, at the last of this paper, a ducted propeller of an all-direction propeller is optimized. Through the process, the duct section size parameters which can generate maximum thrust in the same input condition are obtained.