虚拟试验设计中的序贯极大熵方法研究

Sequential Maximum Entropy Approach to Design of Virtual Experiment

针对包含计算耗时仿真模型的虚拟试验预测模型建模过程,提出了一种序贯极大熵试验设计方法。通过调整高斯先验过程协方差函数,在预测模型误差较大区域序贯地增加新试验点,合理权衡试验点数目与预测模型精度。设计数学算例演示了序贯极大熵试验设计方法的效率,并证明优于一步法,适用于虚拟试验预测模型建模。

A sequential maximum entropy approach was proposed to statistical metamodeling of time-consuming simulation programs, The approach sequentially locates new training points in region where prediction error is larger, A simple numerical example demonstrates that the proposed approach is efficient and superior to single-stage method.