Prediction method and application of shale reservoirs core gas content based on machine learning

To improve the recovery of shale gas, it is very important to accurately grasp the gas content of shale reservoirs. Considering the problems of low accuracy, strong local limitation and poor adaptability of seismic data of traditional methods such as empirical formulas and regression fitting. Based on machine learning (ML) algorithms such as support vector regression, decision trees, random forests, BP neural networks and convolutional neural networks, an intelligent prediction method of shale reservoir core gas content based on machine learning was established by using three parameters: P-wave velocity (Vp), S-wave velocity (Vs) and density (RHOB). It was compared with traditional method prediction, core tests and other gas content data, which verified the effectiveness and high-precision characteristics of the method. Additionally, in support vector regression (SVR), decision tree (DT), random forest (RF), BP neural network (BP), convolutional neural network (CNN) and other machine learning algorithms, the support vector regression algorithm was the most stable, robust and accurate. Because it takes the easy-to-obtain “three parameters” as input data and retains the gas content characteristics of core data, it also has strong generalization ability and easy migration advantages, which can be easily extended to gas content prediction of three-dimensional shale reservoirs based on seismic inversion data. The prediction results of this method in the core gas content of the shale reservoir of the Wufeng Longmaxi Formation in the southern Sichuan Basin show that compared with the traditional method, the gas content prediction accuracy based on the machine learning algorithm was higher. Therefore, it can provide method support for shale reservoir target optimization and drilling deployment.