基于自适应动态替代模型的DNAPLs污染地下水修复过程优化设计

Focuses on the challenging problems on DNAPLs-contaminated groundwater remediation process, this project optimizes and analyzes the remediation strategy of the contaminated groundwater by comprehensive applying of multi-phase flow simulation model, optimal orthogonal-array-based Latin hypercube sampling (OOLHS) method, adaptive dynamic surrogate models and nonlinear programming. First, on the basis of multiphase flow numerical simulation model, OOLHS method is adopted to collect initial input-output data which can reflect the input-output relationship of multi-phase flow simulation model. Then the coverage and uniformity degree of sample points obtained with OOLHS method are appreciated. The initial Kriging surrogate model and the corresponding optimization model are constructed and solved, and the initial optimal remediation strategy can be obtained. Trust region updated criterion, as well as maximum expected improvement criterion are selected as sample added criterion, and the new samples are added adaptively, the two dynamic surrogate models and the corresponding optimization models based on these two sample added criterions are constructed and solved, and the optimal remediation strategy of DNAPLs-contaminated groundwater can be obtained. At last, the merits and demerits of the dynamic surrogate models are evaluated compared with the static surrogate model. The research of this project can effectively improve the surrogate model accuracy and optimization result reliability, and enrich the theory and methodology for the optimization design of DNAPLs-contaminated groundwater remediation process.

本项目针对DNAPLs污染地下水修复问题展开研究，通过多相流数值模拟、最优正交拉丁超立方抽样、自适应动态替代模型及非线性规划等方法的综合运用，对DNAPLs污染地下水的修复过程进行优化设计。在建立多相流模拟模型的基础上，选用最优正交拉丁超立方法获取反映模拟模型输入-输出关系的初始样品数据集，分析其对抽样空间覆盖度和样品点散布均匀性的影响；应用克里格法建立初始替代模型及基于初始替代模型的优化模型并求解，获得修复方案的可能最优解。然后，以信赖域更新准则和最大化改善期望准则作为样品点增加准则进行自适应抽样，分别建立基于两种自适应抽样的动态替代模型及优化模型并求解，得到DNAPLs污染地下水修复过程的最优设计方案。与静态替代模型对比，分析自适应动态替代模型的可行性和优缺点。本项目的研究能有效提高替代模型对模拟模型的近似精度及优化结果的可靠性，丰富DNAPLs污染地下水修复过程优化设计的理论和方法。

在建立多相流模拟模型的基础上，选用最优正交拉丁超立方法获取反映模拟模型输入-输出关系的初始样品数据集，与已有的拉丁超立方抽样方法相比，最优正交拉丁超立方抽样能够同时考虑抽样结果的空间填充性方面及正交性，其获取的样品点更好的代表了整个抽样空间。并且基于最优正交拉丁超立方抽样的静态替代模型对模拟模型的精度也大大提高。以信赖域更新准则及最大化改善期望准则作为样品点增加准则进行自适应抽样，分别建立基于两种自适应抽样的动态替代模型及优化模型并求解，得到DNAPLs污染地下水修复过程的最优设计方案。从替代模型的精度、迭代次数及优化模型的可靠性等方面对静态替代模型和两个基于不同准则的自适应动态替代模型的优缺点进行对比分析。从中甄选出了最优的替代模型——基于信赖域更新准则的动态替代模型，并根据基于此替代模型的优化模型得到了最优修复方案。本项目的研究能有效提高替代模型对模拟模型的近似精度及优化结果的可靠性，丰富DNAPLs污染地下水修复过程优化设计的理论和方法。

内容获取失败，请点击重试