Abstract: The feasibility of design optimization based on metamodel reliability highly depends on the accuracy of metamodel.The selection of correlation functions is vital to the accuracy of Kriging model in modeling process.The existing Kriging modeling methods do not account for any uncertainty in the form of the correlation function,which lead to an undesirable accuracy and robustness metamodel.To conquer the issue,a new Kriging ensemble modeling approach that considered the uncertainty in the model form of the correlation function was proposed.Before implementing the model selection,the expected improvement sampling strategy was used to construct different candidate Kriging models.The 0-1 programming was adopted to the model selection where redundant Kriging models were eliminated before constructing a new Kriging ensemble model.Then,the selected candidate Kriging models were combined as a final model.The proposed Kriging ensemble modeling approach was applied to a bridge-type amplification mechanism to illustrate its effectiveness.The results revealed that the proposed Kriging ensemble modeling method not only improved predictive accuracy,but also enhance the predictive robustness.

Key words: Kriging model, reliability-based design optimization, combinational model, 0-1 programming

摘要： 基于代理模型可靠性优化设计高度依赖于代理模型的精确性。在Kriging模型建模过程中,相关函数的选择往往影响Kriging模型精度。针对相关函数选择不确定情形下Kriging组合模型建模精度和稳健性偏低的难题,提出一种0-1规划模型筛选策略的Kriging组合建模方法。首先,依据期望提高自适应加点准则构建不同相关函数的Kriging模型作为候选模型;其次,通过0-1规划模型筛选策略对候选模型进行优化筛选以剔除预测性能不佳的候选模型;最后,加权平均筛选出的Kriging模型获取最佳Kriging组合模型。以复杂精密机械产品为研究载体验证了Kriging组合建模方法的有效性。研究结果表明,Kriging组合建模方法比单个Kriging模型的预测性能更加精确和稳健。

关键词: Kriging模型, 可靠性优化设计, 组合模型, 0-1规划