Optimization Strategy Using Dynamic Metamodel Based on Trust Region and Biased Sampling Method

Combining a trust region method with a biased sampling method, a novel optimization strategy (TR-BS-KRG) based on a dynamic metamodel is proposed. Initial sampling points are selected by a maximin Latin hypercube design method, and the metamodel is constructed with Kriging functions. The global optimization algorithm is employed to perform the biased sampling by searching the maximum expectation improvement point or the minimum of surrogate prediction point within the trust region. And the trust region is updated according to the current known information. The iteration continues until the potential global solution of the true optimization problem satisfied the convergence conditions. Compared with the trust region method and the biased sampling method, the proposed optimization strategy can obtain the global optimal solution to the test case, in which improvements in computation efficiency are also shown. When applied to an aerodynamic design optimization problem, the aerodynamic performance of tandem UAV is improved while meeting the constraints, which verifies its engineering application.