Improved modeling of channel prediction based on gray relational analysis and a support vector machine: a case study on the X pilot area in the Daqing oilfield in China

Considering the complex reservoir conditions and rapid changes in lithological facies, it is difficult to predict the channel distributions in the Heidimiao oil layer in the X pilot area of the Daqing oilfield. To address this problem, a model for fluvial reservoir prediction under complex geological conditions is established by combining gray relational analysis (GRA) and a support vector machine (SVM). Attribute selection is firstly processed based on 2D forward modeling. A predictive model of the main channel combining GRA and SVM methods is then built using the selected attributes as inputs. The predictive pay thickness is our proposed model is well validated with the realistic pay thickness data interpreted from 18 wells, and all the relative errors are within 10%. Channel predictions from our proposed models also confirmed the accuracy based on historical oil production.