Sequential experimental design for metamodeling based on stochastic Kriging
In test and evaluation of complex systems,the number of tests is limited due to constraints in time and cost.Thus,proper designs are needed for test point and sample size.Sequential design is suitable for high-cost tests but its performance is determined by the design criterion and method utilized.To address this problem,a sequential design strategy is proposed based on the stochastic Kriging.For test point design,a design criterion is established by estimating components of the expected prediction error while centered-L2 discrepancy is introduced to balance exploration and exploitation.For sample size allocation,an integer programming problem with an integrated mean squared error as an objective function is converted into a multistep decision process,and an approximate-dynamic-programming-based solving method is proposed with the value function approximated based on the solution of the slack problem.Finally,the proposed sequential design method is applied to a numerical simulation and a real-world flight simulation turntable experiment respectively to verify its validity.
sequential designstochastic Krigingmetamodelingdesign of experimentssurrogate modelapproximate dynamic programming
NETL
NSTL
万方数据
