Geochemical Artificial Intelligence Tool for Enhanced Water Management

Abstract

Abstract Produced water from crude oil and gas producing zones is geochemically heterogeneous due to the input from diverse underground sources and/or operation-related fluids. The present work targets developing and implementing an innovative numerical tool to automatically predict the type and provenance of recovered produced water from exploration and production. Machine learning-based tools was assessed for the automated classification of produced water types in oilfields with a differentiation of formation water from operation-related fluids. Multiple algorithms in machine learning were tested on hundreds of training and testing datasets of produced water samples to identify methods with the highest performance rates. A combination of the supervised XGBoost algorithm with semi-supervised self-training methods resulted in the two best-performing algorithms for all metrics (accuracy, precision, recall) with an accuracy above 90%. The developed software allows the automated recognition of representative formation water samples, and its differentiation from drilling or operation-related fluids, such as mud filtrate, completion brine, supply water, condensate water, and mixed water types. The algorithm was successfully tested for the classification of hundreds of produced water samples from different fields and reservoirs to define accurate formation water properties for several wells and lithological units. In general, the majority of produced water samples resulted in being post-testimonies from drilling, workover, and production-related activities. Enhanced water management by processed geochemical data can directly contribute to a potential reduction in drilling lost-time, well cleanup time, and reducing the risk of drilling dry holes. Additionally, well control can be optimized by identifying water inflow zones, and the amount of required bottomhole samples can be reduced by identifying nonrepresentative sampling intervals. Furthermore, geochemical fingerprinting of formation water as dominant reservoir fluid can provide analogous clues on the source rock, migration, trapping, and alteration of hydrocarbons.