Shear Wave Velocity Prediction of Carbonate Reservoirs Based on CatBoost

Due to the diverse lithology and complex pore structure of the carbonate reservoirs, the conventional shear wave (S-wave) velocity prediction method is insufficient in mining information, resulting in low accuracy. Therefore, CatBoost is proposed to predict S-wave velocity in complex carbonate reservoirs. Different from the point-to-point learning of continuous variables by traditional machine learning, CatBoost can discover the mixed variables composed of continuous wireline logs and discrete local sequence features, mine the internal relationship between wireline logs and S-wave velocity, build a high precision S-wave velocity prediction model. In the research of the carbonate reservoir in Sudong area of Sulige gas field, sixteen logging parameters sensitive to S-wave velocity are selected, and local wireline logs and lithologic sequence information are introduced to construct a prediction model based on CatBoost. Compared with traditional machine learning, such as SVM, BP neural network, and conventional Xu-Payne petrophysical model, the RMSE of the S-wave prediction model based on CatBoost is only 2.87μs/m but the accuracy reaches 97.59%. The results show that the S-wave prediction model based on CatBoost is more suitable for actual geological distribution and provides a new idea for further research on complex reservoirs.