项目名称: 基于贝叶斯稀疏约束的孔隙介质弹性波方程反演方法研究
项目编号: No.41304092
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 天文学、地球科学
项目作者: 傅红笋
作者单位: 大连海事大学
项目金额: 25万元
中文摘要: 本项目针对流体饱和多孔隙介质弹性波方程反问题,基于贝叶斯理论、稀疏约束和马尔科夫链蒙特卡罗方法,开展以快速、精细反演为目标的反演理论和反演方法研究。在贝叶斯理论框架下,将模型参数在某变换域上的稀疏表示作为先验信息,并利用l1-范数将其描述为解的先验概率分布,基于层次贝叶斯公式,将正则化参数、噪声水平、待反演参数均作为随机变量,以解决正则化参数难以选取和解的不确定性难以量化的问题。为了实现后验概率密度函数的高效率采样,设计求解基于l1-类先验的两种有效MCMC方法:延迟拒绝-自适应Metropolis采样算法和超松弛单分量Gibbs抽样算法,实现对不连续介质参数的高分辨率反演。结合小波多尺度方法,进一步实现全局收敛的自适应多尺度贝叶斯稀疏约束反演算法,实现对尖、角边界的有效识别。本项目研究的是地震勘探波形反演中的难题,同时方法理论具有普遍适用性,对于其它领域中的各种反问题具有重要的参考价值。
中文关键词: 孔隙介质;全波形反演;贝叶斯推断;稀疏约束正则化方法;马尔科夫链蒙特卡罗方法
英文摘要: The aim of this project is to study the inversion theory and fast, fine inversion methods for the inverse problems of wave equation in fluid saturated porous media based on Beyesian theory, sparsity-constraints and Markov chain Monte Carlo sampling methods. We present a Bayesian framework for reconstructing medium parameter from full waveform data by imposing sparsity on the distribution of the parameters in a sparse transform basis through l1-norm prior distribution. In hierarchical Bayesian modeling, all unknowns are treated as stochastic quantities with assigned probability distributions. The hierarchical formulation which determines automatically the regularization parameter and noise level together with the inverse solution will be adopted. To draw samples for the posterior probability distribution in a fast and efficient way, we will develop and examine two new implementations of Delayed Rejection Adaptive Metropolis schemes and single component Gibbs MCMC sampler for sparse relying on l1-norm. Following the work we have accomplished, we will further realize the global convergence of the adaptive multi-scale Bayesian sparse constraint inversion algorithm for discontinuous medium parameter.The proposed framework can be applied to various inverse problems in other fields.
英文关键词: Porous media;Full waveform inversion;Bayesian inference;Sparsity constraint regularization;Markov Chain Monte Carlo method