AI-Drivenriven Reservoir Characterization: Case Studies from the UAE

Abstract The demand for automation has risen considerably in the oil and gas industry – across both Upstream and Downstream segments. This is where Artificial Intelligence (AI)-driven processes come into limelight. The present work showcases the power of AI in characterizing and comprehending oil reservoir rock properties. In an era when integrated analysis of well log data is carried out mostly manually and requires days to complete, a fully automated machine-learning-driven-approach facilitates the work in few hours – if not minutes. In petrophysics, Well-Bore Image (WBI) logs need thorough analysis due to their high geological content, making them more time-consuming to interpret than other logs. In this paper, we demonstrate examples from three different geological Formations, where AI was deployed to analyze WBI data. Log datasets (including WBI) were gathered from different Carbonate Formations of differing rock characteristics - spanning Cretaceous to Jurassic age. The wells targeted were deviated to horizontal producers and injectors that had comprehensive data acquisition. Machine-learning was deployed to carry out WBI-based structural and textural heterogeneity analysis. AI-driven key deliverables included fracture/fault characterization, quantifying diagenetic porosities and extraction of rock permeability indices. AI-driven analysis saves time (and cost) while sustaining quality. What usually takes days or even weeks is now accomplishable in hour or even minutes. The results were able to elucidate better the production and injection flow profiles in comparison to conventional outputs. This in turn helped to comprehend and predict reservoir behavior in a much more relevant manner. Having said so, it is also vital to note here that the AI machine learning is still a continuous (iterative) process and will be tested for several other target Formations of varying geology and age. The refined workflow incorporates supervised machine learning (ML) and computer vision to analyze electrical well-bore images at each depth, enhancing texture analysis and image porosity evaluation for detailed permeability prediction. Utilizing finite element Darcy experiments, the system simulates permeability variations along the wellbore under diverse boundary conditions. This AI-driven approach assesses the impact of different pore structures on overall permeability, applying machine learning algorithms to classify and quantify these effects with high accuracy.