f Leveraging Machine Learning to Improve Subsurface Interpretation Workflows

Machine learning (ML) uses advanced computer algorithms to better understand the relationships between large and multi-variable datasets. Machine learning is useful to geoscientists because it solves two problems: interpreting large volumes of data and understanding the relationship of various types of data at once.

One of the major tasks for geologists is to interpret/pick multiple formation tops from well logs over many hundreds to thousands of wells in a basin. It is very expensive and time consuming as well. Our method uses CNN (Convolutional Neural Network) to learn salient patterns in the well logs and then extrapolates to unseen logs. One of the main advantages of this algorithm is it gets significantly better with more training data.

Petrophysical evaluation of well logs is an essential tool for reservoir description and hydrocarbon resource evaluation. Well logs acquired in bad borehole conditions often exhibit poor quality which needs to be edited or removed. Complete suites of logs are required for many petrophysical models. When the curves are deficient, they must either be acquired over the interval of interest to enable running a complete reservoir characterization workflow or estimated through the calculation of pseudo logs. Re-running of a well log tool is often expensive, so geoscientists and engineers must rely on pseudo log generation methods. Machine learning algorithms are useful to extract patterns and structures from historical data to predict the missing data. The advantage with ML techniques is that they can handle many logs of different types to glean the inter-relationships among the log and reservoir properties. ML algorithms use the well log data from the adjacent wells to learn the underlying behavior of the system without prior knowledge of the nature of relationships between well log data points. We present the machine learning methods that are being developed to improve well log analysis workflows.