局部参数化子域POD-TPWL在大规模油藏历史拟合中的应用
Subdomain POD-TPWL with Local Parameterization for Large-Scale Reservoir History Matching Problems
摘要:
我们最近提出了一种有效的子域POD-TPWL(Xiao et al,2018),用于历史匹配,而无需模型入侵。从计算的角度来看,局部参数化的参数分别定义在每个子域是非常有吸引力的。在这项研究中,我们提出了一个这样的局部参数化,通过集成主成分分析与区域分解独立地表示在每个子域的低序参量子空间内的空间参数场。这种局部参数化允许我们同时扰动每个子域中的参数,并且所有这些扰动的影响可以仅用几个全阶模型模拟来计算。为了避免在相邻子域的边界处的非光滑性,最优局部参数被投影到全局参数化上。子域POD-TPWL与局部参数化的性能进行了评估,通过两个案例测试的修改版本的SAIGUP模式。这种局部参数化具有很高的可扩展性,因为训练模型的数量主要取决于每个子域中的局部参数的数量,而不是底层全阶模型的维度。激活更多的子域会导致更少的局部参数模式,并使我们能够运行更少的模型模拟。对于一个大规模的案例研究,为了优化282个全局参数,LP-SD POD-TPWL只需要90个保真度模拟。与基于有限差分的历史匹配和具有全局参数化的子域POD-TPWL(其中在整个域上定义参数)相比,CPU成本降低了几个数量级的因子,同时保持合理的精度。
英文摘要:
We have recently proposed an efficient subdomain POD-TPWL (Xiao et al, 2018) for history matching without the need of model intrusion. From a computational point of view, a local parameterization where the parameters are separately defined in each subdomain is very attractive. In this study we present one such local parameterization through integrating principle component analysis with domain decomposition to independently represent the spatial parameter field within low-order parameter subspaces in each subdomain. This local parameterization allows us to perturb parameters in each subdomain simultaneously and the effects of all these perturbations can be computed with only a few full-order model simulations. To avoid non-smoothness at the boundaries of neighboring subdomains, the optimal local parameters are projected onto a global parameterization. The performance of subdomain POD-TPWL with local parameterization has been assessed through two cases tested on a modified version of SAIGUP model. This local parameterization has high scalability, since the number of training models depends primarily on the number of the local parameters in each subdomain and not on the dimension of the underlying full-order model. Activating more subdomains results in less local parameter patterns and enables us to run less model simulations. For a large-scale case-study in this work, to optimize 282 global parameters, LP-SD POD-TPWL only needs 90 fidelity simulations. In comparison to the finite-difference based history matching, and subdomain POD-TPWL with global parameterization where the parameters are defined over the entire domain, the CPU cost is reduced by a factor of several orders of the magnitude while reasonable accuracy remains.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。