Research On Automatic History Matching Of Fluvial Reservoir Based On Geological Knowledge Dictionary

The fluvial reservoir has the characteristics of large oil layer span,wide oil-bearing area distribution and strong heterogeneity between reservoirs,and long-term water injection development has made the geological conditions of this type of reservoir more complicated.Accurate identification of the geological features of the river facies has become a prerequisite for well-designed oilfield development programs and enhanced oil recovery.Reservoir numerical simulation and automatic history matching techniques are one of the commonly used methods for identifying reservoirs today,and reasonable reservoirs are selected.The automatic history matching method helps to obtain more accurate reservoir characteristics information in the reservoir.This paper uses the ensemble Kalman filter algorithm to achieve this process.This algorithm is a widely used intelligent algorithm in the field of automatic history matching.However,in dealing with strong heterogeneous reservoirs or assimilation data,it is easy to appear filter divergence and dimensionality disaster.In order to solve this problem,the paper proposes two kinds of problems.The solution is an ensemble smoothing algorithm with multiple data iterations,and an improved Kalman filtering algorithm based on sparse learning.The ensemble Kalman filter algorithm requires the reservoir model to be restarted and considers the reservoir dynamics during the assimilation process.Ensemble smoothing is a viable alternative.Because the ensemble smoothing algorithm only computes a single global update,its inversion and fitting effects need to be improved.The multiple iterations of the ensemble smoothing algorithm uses the improved covariance matrix to perform multiple data assimilation observations in order to improve the results of the ensemble smoothing algorithm,which is based on the equivalence between single and multiple data assimilation in the case of linear Gaussian.The restart of the reservoir model is avoided and the reservoir model is subject to the material balance equation.In addition,considering the characteristics of the number,direction and connectivity of rivers,different types of facies reservoirs' geological knowledge bases are constructed,which enhances the inversion effect.Aiming at the problem of high reservoir model dimension,a geological parameter extraction method based on sparse optimization is proposed and combined with dictionary learning theory.By comprehensively considering the prior geological model and observation data,the static geological parameters in the Kalman filter are transformed into sparse vectors,which can effectively extract the characteristic information of the permeability field,and at the same time greatly reduce the parameter dimension and enhance the algorithm.Compared with the traditional ensemble Kalman filter,the ensemble Kalman filtering algorithm based on sparse learning can track the dominant channel changes in real time according to the observed data changes,thus achieving the tracking and inversion process of the permeability field.