31 Jul 2024
09:00 Doctoral defense IC3 Auditorium
Characterization of Brazilian Pre-Salt Reservoirs Based on Image Analysis and Machine Learning
Leticia da Silva Bomfim
Advisor / Teacher
Hélio Pedrini - Co-supervisor: Alexandre Campane Vidal
Brief summary
Carbonate reservoirs have heterogeneity as one of their main characteristics. This particularity, when linked to the analysis and characterization of wells and reservoirs, adds challenges to the interpretation and definition of a concrete geological model. Furthermore, the entire process of conceptualizing a reservoir includes facing geological uncertainties related to its physical structure, as well as economic uncertainties related to cost and investment. In Brazil, the Santos Basin presents itself as a major producer in the oil and gas sector, which brings together great efforts to understand and model the field. However, its geological formation derived from carbonate rocks presents several challenges involving its highly heterogeneous constitution. Thus, strategies that allow deepening and producing generalizations of these environments enable optimization and greater success in the exploration of a reservoir. To automate and improve the procedures involved in this task, several approaches in the field of computer vision have been applied over the years, and can be observed from data acquisition to the final modeling of reservoirs. Methodologies developed for this geological characterization demonstrate a strong synergy between image processing techniques and machine learning. This mutualistic collaboration contributes to the advancement of both technologies, collectively providing a variety of solutions to the diverse challenges encountered in geological reservoir modeling. Therefore, through this research we seek to contextualize the use of image analysis and machine learning in the characterization of reservoirs, as well as to present in a practical way the application of this resource in three important tasks of the geological modeling workflow: The classification of seismofacies through convolutional networks, the identification of flaws in seismic data using Transformers, and the improvement of noisy well images to enable data analysis and structure identification, through image processing techniques. All of these processes were evaluated quantitatively and qualitatively, and corresponded positively to the methods applied. This approach reinforces visual computing as an essential and powerful tool in reservoir characterization, driving accuracy, automation and efficiency, and promoting significant advances in geological exploration and modeling.
Examination Board
Hélio Pedrini IC / UNICAMP
Andre Santanche IC / UNICAMP
Camila Duelis Viana IGc/USP
Ronaldo Cristiano Prati CMCC / UFABC
Emilson Pereira Leite IG/UNICAMP
Guilherme Furlan Chinelatto CEPETRO / UNICAMP
João Paulo Papa FC / UNESP
Moacir Antonelli Ponti ICMC / USP