Reservoir permeability prediction based artificial intelligence techniques

Predicting permeability is a cornerstone of petroleum reservoir engineering, playing a vital role in optimizing hydrocarbon recovery strategies. This paper explores the application of neural networks to predict permeability in oil reservoirs, underscoring their growing importance in addressing traditional prediction challenges. Conventional techniques often struggle with the complexities of subsurface conditions, making innovative approaches essential. Neural networks, with their ability to uncover complicated patterns within large datasets, emerge as a powerful alternative. The Quanti-Elan model was used in this study to combine several well logs for mineral volumes, porosity and water saturation estimation. This model goes beyond simply predicting lithology to provide a detailed quantification of primary minerals (e.g., calcite and dolomite) as well as secondary ones (e.g., shale and anhydrite). The results show important lithological contrast with the high-porosity layers correlating to possible reservoir areas. The richness of Quanti-Elan's interpretations goes beyond what log analysis alone can reveal. The methodology is described in-depth, discussing the approaches used to train neural networks (e.g., data processing, network architecture). A case study where output of neural network predictions of permeability in a particular oil well are compared with core measurements. The results indicate an exceptional closeness between predicted and actual values, further emphasizing the power of this approach. An extrapolated neural network model using lithology (dolomite and limestone) and porosity as input emphasizes the close match between predicted vs. observed carbonate reservoir permeability. This case study demonstrated the ability of neural networks to accurately characterize and predict permeability in complex carbonate systems. Therefore, the results confirmed that neural networks are a reliable and transformative technology tool for oil reservoirs management, which can help to make future predictive methodologies more efficient hydrocarbon recovery operations.