Friction Reducer Advisor: A Real-Time Machine Learning Approach

Abstract In unconventional fracturing, current practices for friction reducer (FR) concentration management lack engineered techniques. The modification of FR typically is using pressure trends and established practices that could lead to high amounts of FR utilization and suboptimal project economics. Here we present a novel approach to address this by incorporating an advisory machine learning (ML) model that could be used in real time. The model is capable of recognizing trends to predict when a change in concentration is ideal. The historical stimulation data from more than 250 stages from 10 different wells is mined and appended with domain fundamentals to generate recommendations for adjusting FR concentrations. An essential step involved feature engineering, enabling the creation of new variables to extract meaningful insights from the data. As the data are unlabeled, two methodologies that included unsupervised learning and domain logic were compared to create the classes. In this study we present a novel workflow for developing an FR advisor that provides categorical recommendations to increase, decrease, or hold FR concentrations during hydraulic fracturing operations. The model is designed to selectively capture data points that exhibit a direct pressure-FR relation while disregarding points that have other variables influencing pressure. A classification model exhibited the highest accuracy of around 94%. Implementing the advisor recommendations can potentially yield significant FR savings ranging from 3.4% to 21.1% of the total FR consumed in each stage while maintaining an optimal stimulation. The potential cost savings ranges from USD 12,000 to USD 42,000 per well while manufacturing and transportation emissions reduction can be up to 182.3 t carbon dioxide equivalent (CO2e) per well. A detailed benefit analysis was conducted to address concerns regarding formation damage, additional wear on pumping equipment, and associated emissions. Additive management lacks real-time adaptability: most treatments focus only on achieving the designed rate and avoiding potential issues, leading to avoidable FR volume utilization that results in a waste of materials. This innovative FR advisor model is useful for fracturing engineers and project managers to enhance operational efficiency while promoting a more sustainable and cost-effective stimulation. The approach can be easily extended to other fracturing materials beyond FR. While operational oversight remains essential, the advisor provides actionable insights with suggestions, leading the way to an enhanced efficiency, reduced material waste, and environmental benefits.