근접점 교차검증을 사용한 국부 오차 평가 기반 앙상블 메타모델

In general engineering design process, physical experiment or computer simulation is repeatedly required and it causes increase of cost for design. To reduce the cost, many researches on the using a meta-model instead of physical experiment or computer simulation have been developed. The polynomial regression, kriging, radial basis function, and support vector regression are widely used in engineering field, and the accuracy of these meta-models are depend on the trend of the true response and the design of experiment. Thus, it is difficult to select a meta-model suitable for the true response. To overcome the shortcoming, researches on the ensemble of meta-models which combining various stand-alone meta-models have been developed. In this study, we proposed v-Nearest Neighbors Points Cross-validation for meta-model validation and ensemble of meta-models based on the proposed validation technique. Test result show robustness and accuracy of the proposed ensemble of meta-model.