【摘要】 针对变可信度近似模型普遍聚焦于单一输出场景建模，在解决多输出问题时人为割裂了输出之间的相关性，导致有用信息丢失的问题，提出了一种变可信度多输出近似模型，通过多输出高斯过程捕获输出间的潜在相关性，并采用变可信度近似建模框架融合多精度数据，从而有效处理多精度数据下的多输出预测问题．通过三个数值算例和超材料隔振器性能预测案例验证了提出的模型方法的有效性，并与目前两种主流的变可信度近似建模方法对比，结果表明：所提出模型在多输出预测问题上整体性能优于现有方法，在超材料隔振器性能预测中，所提出方法较现有变可信度近似模型在全局精度和局部精度方面均得到显著改善．更多还原

【Abstract】 Aiming at the problem that current variable-fidelity surrogate model(VFSM) approaches focused on the modeling of a single output scenario,and the latent correlation across outputs was cut off artificially when solving multi-output problems,leading to the loss of some useful information,a variable-fidelity multi-output surrogate modeling approach was proposed. The latent correlation across outputs was captured by multi-output Gaussian process,and the variable-fidelity data were fused by VFSM framework. Hence,the proposed approach could deal with variable-fidelity multi-output prediction problems efficiently. The effectiveness of the proposed model was verified using three numerical examples and an engineering case in the performance predictions of a metamaterial vibration isolator,and the proposed approach was also compared with two mainstream VFSM approaches.Results show that the proposed approach shows a better overall performance compared with the existing approaches in the predictions of multiple outputs,and in the performance predictions of the metamaterial vibration isolator,the proposed method demonstrates substantial enhancements in both global accuracy and local accuracy compared with the existing VFSM approaches.更多还原