摘要
在水电工程坝基三维离散裂隙网络(Discrete Fracture Network,DFN)随机建模中,关键在于对裂隙的倾向、倾角、开度等几何多参数的多维联合分布估计。然而,现有基于典型分布的DFN模拟方法存在缺乏考虑裂隙参数之间的相关性,且难以实现对裂隙参数多维联合分布的高精度概率密度估计的问题。针对上述问题,本文提出一种改进的自回归流模型——密度峰值聚类自回归流(Density Peak Clustering Autoregressive Flow,DPCAF),通过采用高斯混合分布与DensityPeak聚类算法改进标准化特征空间的基础分布,弥补自回归流在分布估计的过程中难以考虑裂隙优势分组的不足,提高对于多峰的联合分布的拟合能力;进一步提出基于DPCAF模型的裂隙网络多参数模拟方法,考虑裂隙几何参数之间的相关性,实现其多维联合分布的精确极大似然估计与采样。工程应用结果表明,DPCAF模型相比于现有基于典型分布的方法能够更好地拟合复杂的多参数联合分布,且具备能够建立裂隙几何参数关联关系的优势,从而保证了DFN模型的可靠性。
In stochastic modeling of 3D discrete fracture network(DFN)in dam foundations,the key lies in the estimation of multidimensional joint distribution of geometric parameters of fractures including dip,dip direction,aperture etc.However,existing DFN modeling methods are based on classic distributions,which ignore the correlations among parameters and cannot realize the high precision probability density esti⁃mation of the multidimensional joint distribution of fracture parameters.To solve those problems,this paper proposes an improved autoregressive flow model——Density Peak Clustering Autoregressive Flow(DPCAF),in which the Gaussian mixture distribution and DensityPeak clustering algorithm are used to modify the base distribution in normalized feature space.This model makes up for the deficiency of autoregressive flow which lacks consideration of fracture dominant partitioning in distribution estimation,and enhances the fit⁃ting ability to multi-modal joint distributions.Furthermore,a method of multivariate parameter simulation of 3D fracture network using DPCAF model is proposed,which takes the correlations among fracture parame⁃ters into account,and can realize the accurate maximum likelihood estimation and sampling from the multi⁃dimensional joint distribution.Engineering application shows that the DPCAF model can better fit the com⁃plex multidimensional joint distributions of fracture parameters compared with classic distribution-based meth⁃ods,and have the advantage of construct the correlation structures among fracture parameters,which fur⁃ther ensure the reliability of DFN models.
作者
张亦弛
吕明明
关涛
王佳俊
余佳
任炳昱
ZHANG Yichi;Lü Mingming;GUAN Tao;WANG Jiajun;YU Jia;REN Bingyu(State Key Laboratory of Hydraulic Engineering Simulation and Safety,Tianjin University,Tianjin 300072,China)
出处
《水利学报》
EI
CSCD
北大核心
2021年第5期565-577,共13页
Journal of Hydraulic Engineering
基金
国家重点研发计划项目(2018YFC0406704)
国家自然科学基金雅砻江联合基金项目(U1765205)
国家自然科学基金项目(51879186)。
关键词
水电工程
坝基岩体
离散裂隙网络
多参数模拟
自回归流模型
hydropower engineering
rock mass of dam foundation
discrete fracture network
multivariate parameter simulation
autoregressive flow model
作者简介
张亦弛(1991-),博士生,主要从事水电工程三维地质建模研究。E-mail:zhangyichi@tju.edu.cn;通讯作者:关涛(1989-),博士,副教授,主要从事水利水电工程建设过程仿真与控制研究。E-mail:tao.guan@tju.edu.cn。
