基于最大超体提升期望的代理模型多目标优化补充采样策略研究

Research on Maximum Hypervolume Expected Improvement Sampling Method for Surrogat-Model-Based Multi-objective Optimization

使用代理模型进行多目标优化时,代理模型精度决定了优化结果的可信度。要提高代理模型精度,通常需要补充采样。本文提出了一种基于最大超体提升期望的采样策略,通过切片分区计算方法计算得到超体体积,基于克里金代理模型提供预测置信度的特点,采用最大超体提升期望策略(MAXHVEI)进行补充采样,与其他传统补充采样策略相比,本文提出的MAXHVEI补充采样策略有效的减少补充采样个数,有效缩短工程实际中多目标优化问题的求解时间。

When using surrogate model to solve multi-objective problem,the reliability of optimization result is determined by the accuracy of the surrogate model itself.While to improve the accuracy of the surrogate model,added sampling is necessary.A sampling strategy based on the maximum super volume expected improvement(MAXHVEI)is proposed,in which the slice section strategy is integrated.By taking advantage of the Kriging surrogate model which can provide its predicting results together with confidence,the proposed MAXHVEI method is proved to help decreasing the sampling number in gaining a high-accuracy surrogated model and to help saving time in practical multi-objective engineering problem,when compared to other tradition sampling method.