A Method for Predicting Formation P-wave Velocity by Integrating Geological Modeling and Machine Learning

Accurate prediction of formation P-wave velocity is crucial for various stages throughout the entire lifecycle of oil and gas exploration and development. To address the limitations of traditional P-wave velocity prediction methods, such as the insufficient lateral control range based on logging data, low vertical prediction accuracy from seismic data, and the inability to achieve advanced predictions in exploration areas, a synergistic and complementary approach that integrates logging data and seismic data was proposed to reduce uncertainty. A machine learning sample library was constructed using a 3D geological model and sonic logging data. Three ensemble learning models (random forest, gradient boosting decision tree, and light gradient boosting machine) were employed to identify the optimal strong learner suitable for predicting formation P-wave velocity. Case studies were conducted on data completion and advanced prediction of P-wave velocity for both drilled and to-be-drilled wells. The research results indicate that the gradient boosting decision tree model outperforms both the random forest and light gradient boosting machine models in prediction accuracy and generalization ability for formation P-wave velocity, with a coefficient of determination of 0.8895 and a mean absolute percentage error of only 9.39%. The selected optimal strong learner can intelligently complete the P-wave velocity for drilled wells with an average error of only 8.81%. Furthermore, it can also predict the P-wave velocity of to-be-drilled wells in advance. Rock mechanics parameters and safe density windows for drilling fluids can be obtained by further integrating the relevant theoretical models. This research provides a novel approach for predicting P-wave velocity and is significant in promoting the integration of geology and engineering.