Formation pressure prediction in carbonate reservoirs based on machine learning
Formation pressure is a crucial parameter in the process of oilfield development,involving adjustments such as development methods and production allocation and injection.However,obtaining formation pressure typically requires shut-in wells for pressure buildup tests,which are cumbersome.Numerical simulation methods are also labor-intensive and are time-consuming,while existing empirical formula methods are not well-suited for complex operational regimes in carbonate reservoirs.In this study,a data-driven prediction model for formation pressure is developed using a combination of correlation calculation,principal component analysis,elite strategy genetic algorithm,and support vector regression(SEGA-SVR).The SEGA-SVR model achieved high performance metrics,with a coefficient of determination(R2)of 0.97 and a root mean square error(RMSE)of 0.04 on the training set,and R2 of 0.95 and RMSE of 0.05 on the testing set.The model also demonstrated good performance on validation wells in neighboring areas.Compared to traditional SVR models,the SEGA-SVR model showed significant improvement in performace.Additionally,it outperformed other machine learning models overall.The research results show that the SEGA-SVR model can predict real-time formation pressure without the need for well shut-in,and the parameter tuning using genetic algorithm is efficient.The data-driven approach can allows for better adaptation to complex situations.Moreover,the model exhibits good generalization and stability,providing a new method for predicting formation pressure in carbonate reservoirs.
万方数据
维普
