Innovative Machine Learning Approach at Unconventional Production Prediction: The Type-Curve Optimizing Geostatistical Array (TOGA)
Presented by Shane J. Prochnow at the February Meeting of the Permian Basin Section-SEPM
Tuesday, February, 16, 2021
ABSTRACT: Several years ago, Chevron introduced an innovative data analytic workflow called the Type-Curve Optimizing Geostatistical Array (TOGA) (U.S. Patent No. 62/564357). TOGA ties unconventional reservoir and well completion characteristics to historical production and makes robust predictions of well performance. TOGA is built on a foundational technology that utilizes random forest machine learning geostatistically optimized for use on spatially autocorrelated and clustered sample data. Random forest methods are preferred for unconventional systems because they capture complex, multidimensional and non-linear interactions in noisy databases. The random forest methods are also easy to hyper tune to generalize (not overfit), providing confidence that the correct primary drivers are accounted for in production predictions. TOGA is part of a portfolio of other machine learning and physically-based reservoir engineering tools used by Chevron to reduce uncertainty in reservoir performance prediction and type curve generation. TOGA is designed to quantify expected recovery across a large expanse of a hydro-carbon producing basin given the historical relationship between many dimensions of production, reservoir properties and completion practices. The method provides the requisite output to construct predicted, synthetic type curves for any location within a basin. The workflow generates performance prediction maps by applying the random forest multivariate relationships to grids of the key reservoir predictor variables. These production prediction maps are stacked to form an array of locations that each have a unique expected production profile through time. These stacked cumulative production predictions are transformed arithmetically into incremental production rates and can be subjected to traditional decline curve analysis to determine b-factors, EUR, and evaluate the spatio-temporal production heterogeneity of a region. TOGA output may be used to calibrate existing type curve workflows, define reservoir sweet spots, establish reservoir continuity, and predict ultimate recovery for business and economic planning. Data science and machine learning approaches have revolutionized Chevron’s understanding of its unconventional interests, especially in the Permian Basin. The Permian’s unconventional plays were once thought to be relatively unpredictable, highly variable, and having little connection to reservoir properties. TOGA and other proprietary data analytic technologies have enabled the identification of key reservoir performance drivers for each unconventional target zone under development. These drivers are then used to predict well performance through time, routinely within 70%- 85% confidence, providing a key competitive advantage for Chevron.
BIOGRAPHY: Shane J. Prochnow works at Chevron Technology Center in the Subsurface Innovation Lab as a Digital Geology Advisor. Dr. Prochnow has 15 years of industry experience with ExxonMobil and Chevron. He has a Ph.D, MS, and BS in Geology from Baylor University, and is a former Army National Guard Officer. His research interests include unconventionals, reservoir characterization, geostatistics, machine learning, and integrating complex systems.
Shane has been a Reservoir Characterization Advisor for all Chevron Unconventional Business Units, a Development Geologist for Chevron's Mid Continent Asset Development Group. And a Geologic Advisor for the Applied Reservoir Management Team Completions Optimization Group.