Tese e Dissertação

Tese: Applications of machine learning methods in oil and gas for soft sensoring, economic assessment and multiphase flow simulation

Machine learning has seen rapid adoption across diverse sectors—from healthcare and finance to energy—due to its ability to uncover patterns and make predictions from complex datasets. In the oil and gas industry, this work explores three key applications of machine learning: soft sensoring for virtual flow metering, early-stage economic assessment of exploratory assets, and hybrid modeling for multiphase flow simulation. For soft sensoring, system identification techniques combined with current-time data significantly improved predictive accuracy while maintaining model utility for forecasting. In economic assessment, black-box classifiers trained on imbalanced datasets demonstrated feasibility for rapid asset appraisal, with investment data and oversampling strategies yielding notable performance gains, albeit with trade-offs in recall and interpretability. For multiphase flow modeling, hybrid models integrating commercial mechanistic models with data-driven estimators consistently outperformed both standalone physical and black-box models, achieving up to 71% error reduction in pressure gradient prediction. These findings highlight the potential of machine learning to augment traditional engineering workflows, improve decision-making, and address longstanding challenges in oil and gas operations.

Link da defesa:

https://puc-rio.zoom.us/j/95039469339?pwd=ndQ143lEgpXwGeJMa9rsNbEHmkhPbF.1

Ver todos