Automated Well Log Interpretation Through Machine Learning

Abstract Well logs present a concise, in-depth representation of formation parameters. These logs allow interpreters to identify different rock types, distinguish porous from non-porous rocks, and quickly identify pay zones in subsurface formations. The ability to interpret well logs is largely dependent on the interpreter's ability to recognize patterns, past experiences, and knowledge of each measurement. Traditionally, logs were manually corrected for anomalies and normalized at the field scale, which is a time-consuming and often subjective approach. This is especially true for mature fields where log data has been collected from multiple sources. However, the future of petrophysical evaluation is moving towards increased efficiency, accuracy, and objectivity through smart automation. In this paper, we demonstrate the application of machine learning algorithms to automate well-log processing and interpretation of standard log measurements as well as nuclear magnetic resonance (NMR) using data acquired in one of the fields in Iraq. Standard logs such as density, sonic, neutron, gamma ray, etc are classified using machine learning (ML) algorithm into a set of classes that are converted to zones to drive petrophysical interpretation. This novel application of ML algorithm uses cross-entropy clustering (CEC), Gaussian mixture model (GMM), and Hidden Markov Model (HMM) which identifies locally stationary zones sharing similar statistical properties in logs, and then propagates zonation information from training wells to other wells. The training phase involves key wells which best represent the formation and associated heterogeneities to automatically generate classes (clusters), the resulting model is then used to reconstruct inputs and outputs with uncertainty and outlier flags for cross-checking and validation. The model is then applied to predict the same set of zones in the new wells that require interpretation and predict output curves. The main advantage is reducing the turnaround time of the interpretation and eliminating subjective inconsistencies often encountered with standard interpretation approaches. For multi-dimensional data such as NMR, several ML methods such as Parallel Analysis, Factor Analysis, and Cluster Analysis were applied to (a) determine the optimal number of modes to retain in the input NMR T2 distributions, these modes are the underlying poro-fluid constituents affecting NMR data over the entire interval b) decompose T2 distribution into these modes c) compute poro-fluid constituents volumes and cluster it into the same number of groups as the number of factors. This workflow helps to extract maximum information from multi-dimensional NMR data and eliminates the need for any a-priory assumptions, such as T2 cut-offs. We present the results of these methods applied to data acquired across the cretaceous successions in the south of Iraq to speed up the petrophysical analysis process, reduce analyst bias, and improve consistency results between one well to another within the same field.