地下水DNAPLs污染修复多相流模拟的替代模型

应用克里格法、支持向量回归法、核极限学习机法建立多相流模拟模型的替代模型，并应用集对分析法建立组合替代模型，通过将多种替代模型进行对比，分析不同替代模型在地下水DNAPLs污染修复多相流模拟问题中的适用性.结果表明，3种单一替代模型中，克里格模型的精度最高，其次是核极限学习机模型，最后是支持向量回归模型；应用集对分析原理建立集对权组合替代模型，与单一替代模型相比，其模拟模型的逼近精度更高，且提升效果十分显著，平均残差和平均相对误差分别为0.4009%和0.5373%；集对权组合替代模型运行一次仅需1.5s，选择组合替代模型代替多相流模拟模型进行地下水DNAPLs污染的修复方案优选分析，能够大幅减小模拟-优化过程的计算负荷，并保持较高的计算精度.

Surrogate models of multi-phase flow simulation model for DNAPL-contaminated aquifer remediation

Kriging, support vector regression (SVR), kernel extreme learning machine (KELM) were applied to building the surrogate models of multi-phase flow simulation model, and set pair analysis (SPA) was applied to building ensemble surrogate models. The applicability of different surrogate models was analyzed via a comparison study. kriging model was with the highest accuracy, followed by KELM model and SVR model. Compared with Kriging model, set pair weighted ensemble surrogate model significantly improved the approximation accuracy. The mean of residuals and the mean of relative errors between ensemble model outputs and simulation model outputs were only 0.4009% and 0.5373%, respectively. Furthermore, it only took 1.5s to run the set pair weighted ensemble surrogate model. Replacing the simulation model with an ensemble surrogate model considerably reduced the computational burden of the simulation-optimization process and maintained high computation accuracy for optimizing the DNAPL-contaminated aquifer remediation strategy.