文件名称:A Markov Chain Monte Carlo Method for Inverse Stochastic Simulation
文件大小:2.67MB
文件格式:PDF
更新时间:2022-07-10 11:13:26
MCMC
A classical two-stage method to stochastic inverse problems in groundwater and petroleum engineering starts from the generation of a series of independent seed flelds and then calibrates those flelds to inverse-condition on nonlinearly dependent state data from difierent sources, which is known as model calibration or history matching. However, an inherent deflciency exists in this type of method: the spatial structure and statistics are not preserved during the procedure of model calibration and history matching. While the spatial structure and statistics of models may be one of the most important error sources to the prediction of the future performance of reservoirs and aquifers, it should be consistent with the given information just as conditioning to linear data and inverse-conditioning to nonlinear data. In other words, the realizations generated should preserve the given spatial structure and statistics during the procedure of conditioning and inverse-conditioning. Aiming at this problem, a stochastic approach is presented in this study to generate independent, identically distributed (i.i.d) realizations which are not only conditional on static linear data and inverse-conditional on dynamic nonlinear data but also have the specifled spatial structure and statistics.