Synthetic Slowness Shear Well-Log Prediction Using Supervised Machine Learning Models

• Autor(a) principal: Tamoto, Hugo
• Data de Publicação: 2023
• Outros Autores: Contreras, Rodrigo Colnago [UNESP], Santos, Franciso Lledo dos, Viana, Monique Simplicio, Gioria, Rafael dos Santos, Carneiro, Cleyton de Carvalho
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1007/978-3-031-23492-7_11; http://hdl.handle.net/11449/249038
• topic: Forecasting Time-series; Machine learning; Regression models; Synthetic well-logs

The shear slowness well-log is a fundamental feature used in reservoir modeling, geomechanics, elastic properties, and borehole stability. This data is indirectly measured by well-logs and assists the geological, petrophysical, and geophysical subsurface characterization. However, the acquisition of shear slowness is not a standard procedure in the well-logging program, especially in mature fields that have a limited logging scope. In this research, we propose to develop machine learning models to create synthetic shear slowness well-logs to fill this gap. We used standard well-log features such as natural gamma-ray, density log, neutron porosity, resistivity logs, and compressional slowness as input data to train the models, and successfully predicted a synthetic shear slowness well-log. Additionally, we created five supervised models using Neural Networks, AdaBoost, XGBoost, and CatBoost algorithms. Among all models created, the neural network algorithm provided the most optimized model, using multi-layer perceptron architecture reaching impressive scores as R 2 of 0.9306, adjusted R 2 of 0.9304, and MSE less than 0.0694.