| Popis: |
Summary Permeability is one of the key petrophysical parameters of hydrocarbon bearing formations. One of the crucial roles of this parameter is to estimate production rate in oil bearing reservoirs. Permeability is usually measured by core plugs in laboratory. Although the measured value by this method is very precise, it is very expensive and time consuming. Furthermore, core data is not always available for all the wells. Over the last years, regression methods have been widely used to predict permeability in areas with core missing data. Among various regression approaches we selected SVM which works due to SRM and its produced model is less prone to over-fitting problems. Petro-physical properties in heterogeneous reservoirs vary in vertical displacements. So predicting them is difficult due to their abrupt changes. We exploited Electrofacies Analysis using Multi Resolution Graph-Based Clustering to partition formations due to their similarity. We also modified Support Vector Regression to FSVR by giving each data point a membership function to reduce effect of noise and outliers on data.The predicted outputs correlated with core data in the test well. The results show that predicted permeability from FSVR has a notably better correlation with real data compared to correlation outputs from SVR. |
