The simulation and prediction of permeability image are quite significant to the development of oil fields. Because the accuracy of permeability can be improved by integrating conditional data including soft data and hard data

a method based on continuous multiple-point geostatistics (MPS) is proposed to improve the accuracy of permeability in a simulated permeability distribution image. First

filters can reduce the dimensions of permeability images to create a “filter score” space. Second

“filter score” space is partitioned by a two-step partition method to obtain the patterns of each non-empty class and forms a “database” of patterns. Finally

a known pattern is drawn from the above “database” of patterns by a self-defined distance function to achieve a simulated value for an unknown node. The simulated images were compared

showing that the structural characteristics of the permeability image simulated by using both soft data and hard data as conditional data are most similar to those of real data from the training image because of the usage of abundant conditional data.