Using Stacking model fusion to identify fluid in high-temperature and high-pressure reservoir

The difference in logging responses of reservoirs with different fluid properties in Dongfang X gas field is not obvious, and it is especially difficult to determine the bottom limit of the resistivity of different fluids. In addition, it is susceptible to physical factors when using porosity logging curves to identify fluids, making different fluid properties of sample data overlap. A Stacking model fusion method is proposed. It includes various machine learning algorithms (i.e. decision tree, support vector machine, random forest and extreme gradient boosting) and has better effects on fluid identification at high temperature and high pressure. Through 10 fold and cross validations, the algorithms are iteratively optimized to achieve a final optimal model. Compared with a single machine learning algorithm, the Stacking model fusion algorithm can take into account the differences in data observation and training principles of different algorithms, and give full play to the advantages of each model. Tests on real data indicate that the Stacking model can improve prediction accuracy from 87.08% to 92%, compared with the best-performing single model XGBoost. It has a stronger learning ability, and more suitable for fluid identification in high-temperature and high-pressure reservoir. It provides a new idea for building models of logging interpretation. © 2021, Editorial Department OIL GEOPHYSICAL PROSPECTING. All right reserved.