Petroleum Science >2011,??Issue 3:??316-327 DOI: https://doi.org/10.1007/s12182-011-0148-7
Reservoir history matching and inversion using an iterative ensemble Kalman filter with covariance localization Open?Access
文章信息
作者：Wang Yudou and Li Maohui
作者單位：
School of Physics Science & Technology, China University of Petroleum, Dongying, Shandong 257061, China;School of Physics Science & Technology, China University of Petroleum, Dongying, Shandong 257061, China
投稿時間：2010-12-01
引用方式：Wang, Y. & Li, M. Pet. Sci. (2011) 8: 316. https://doi.org/10.1007/s12182-011-0148-7
文章摘要
Reservoir inversion by production history matching is an important way to decrease the uncertainty of the reservoir description. Ensemble Kalman filter (EnKF) is a new data assimilation
method. There are two problems have to be solved for the standard EnKF. One is the inconsistency between the updated model and the updated dynamical variables for nonlinear problems, another is the filter divergence caused by the small ensemble size. We improved the EnKF to overcome these two problems. We use the half iterative EnKF (HIEnKF) for reservoir inversion by doing history matching. During the HIEnKF process, the prediction data are obtained by rerunning the reservoir simulator using the updated model. This can guarantee that the updated dynamical variables are consistent with the updated model. The updated model can nonlinearly affect the prediction data. It is proved that HIEnKF is similar to the first iteration of the EnRML method. Covariance localization is introduced to alleviate filter divergence and spurious correlations caused by the small ensemble size. By defining the shape and size of the correlation area, spurious correlation between the gridblocks far apart is alleviated. More freedom of the model ensemble is preserved. The results of history matching and inverse problem obtained from the HIEnKF with covariance localization are improved. The results show that the model freedom increases with a decrease in the correlation length. Therefore the production data can be matched better. But too small a correlation length can lose some reservoir information and this would cause big errors in the reservoir model estimation.
method. There are two problems have to be solved for the standard EnKF. One is the inconsistency between the updated model and the updated dynamical variables for nonlinear problems, another is the filter divergence caused by the small ensemble size. We improved the EnKF to overcome these two problems. We use the half iterative EnKF (HIEnKF) for reservoir inversion by doing history matching. During the HIEnKF process, the prediction data are obtained by rerunning the reservoir simulator using the updated model. This can guarantee that the updated dynamical variables are consistent with the updated model. The updated model can nonlinearly affect the prediction data. It is proved that HIEnKF is similar to the first iteration of the EnRML method. Covariance localization is introduced to alleviate filter divergence and spurious correlations caused by the small ensemble size. By defining the shape and size of the correlation area, spurious correlation between the gridblocks far apart is alleviated. More freedom of the model ensemble is preserved. The results of history matching and inverse problem obtained from the HIEnKF with covariance localization are improved. The results show that the model freedom increases with a decrease in the correlation length. Therefore the production data can be matched better. But too small a correlation length can lose some reservoir information and this would cause big errors in the reservoir model estimation.
關(guān)鍵詞
-
Half iterative ensemble Kalman filter, covariance localization, reservoir inversion, history matching, fluvial channel reservoir