Well Logs Interpretation Using Machine Learning Workflow

Well logs are used to measure acoustic, nuclear, or conductive properties of the subsurface. In turn, these properties are interpreted based on some physical correlations to compute essential reservoir characteristics such as porosity and water saturation. Most of these physical calculations require a pre-knowledge of the reservoir fluid and rock properties as well as domain knowledge experts and petrophysicsts. This process could also be time consuming and inconsistent due to measurement anomalies and expertise bias. Recent advances in artificial intelligence have produced a paradigm shift in the industry from using traditional physics-based methods to adopting modern data-driven models to reduce physical complexity, improve speed and accuracy. Many research aspects are now focused towards using machine and deep learning models to improve well logs analysis covering many aspects including: detecting anomalies, classification of lithology and automated prediction of reservoir parameters. This paper presents a modern approach of using machine learning modeling workflow for well log interpretation combining both unsupervised and supervised learning technique. The presented model is capable of efficiently predicting several reservoir properties including, water saturation, volume of shale and porosity, without the pre-knowledge of complex physical rock and fluid characteristics.