A New Efficient Adaptive Polynomial Chaos Expansion Metamodel

To address the challenge of the accuracy and efficiency of the metamodel, an adaptive sequential polynomial chaos expansion (ASPCE) metamodel technique is presented. The Latin hypercube sampling (LHS) is used to obtain the initial samples. A new adaptive truncation strategy of polynomial chaos expansion (PCE) is presented for high order PCE, and the parameters are updated by global sensitivity indices got by the Sobol' sensitivity analysis based on the PCE directly. The important terms of PCE are selected by elastic net (EN), and the samples are added according to the combined sequential criterion until the accuracy requirements are satisfied. Thus, by using the presented method, high accuracy model can be constructed by using small number of samples and the global sensitivity indices can be obtained efficiently. At last, three benchmark examples and a numerical example are provided to demonstrate the effectiveness and the efficiency of the presented method.