Characterization and Prediction Method of Oil-Water Relative Permeability Curve Based on XGBoost Algorithm

The oil-water relative permeability curve is an important index that affects the efficiency of water-flood development oilfields. During the exploration and development process, the oil field has accumulated a large amount of coring well data and core experiment data, which provides conditions for the rapid prediction method of the relative permeability curve based on machine learning technology. The first task is to collect and sort out the data of various dynamic and static factors that affect the phase permeability curve, and carry out the pre-processing of the data. A method to describe the shape of the curve by using the improved angle chain code is proposed, and a model to characterize the permeation curve in combination with the conventional feature points of the permeation curve is established, and a learning sample library with 14 sets of characteristic parameters is constructed. then, the paper has established the feature models and included angle chain models based on XGBoost algorithm. Then, a total of five sets of models are trained and optimized. Finally, the general oil-water permeability curve of a single well is quickly obtained through two inversions, and the prediction calculation results are quantitatively evaluated based on the prediction results of the feature points and the included angle chain codes. The comparison between the calculation results of the relative permeability curve based on the data mining technology and the experimental method shows that the prediction accuracy of XGBoost model can reach more than 86%, which meets the oilfield site application requirements. The research shows that this method has high prediction accuracy and strong practicality, and has important guiding significance in reservoir numerical simulation, water drive characteristic analysis, and other aspects.