Completion design is widely considered as a dynamic area of experimentation and development due to its significant impact on production performance in unconventional reservoirs. As the industry moves beyond the paradigm of proppant and base fluid analysis, more and more focus is being given to the composition of chemicals being used in the completion design. This paper focuses on quantifying the true impact of these chemicals used in completion design by using machine learning to solve this multivariable problem and creates value by providing a framework to help completion engineers select the optimum chemicals and aid in fracture treatment design process, using a data-driven approach.
Depending on the shale formation being fractured, a typical fracture treatment job uses a very low concentration of 3-12 additive chemicals. This selection is carried out from thousands of commercially available chemicals and is based on treatment job performance carried out during the modeling phase as well as from field experience. In this work the authors have tried to deconstruct the varied chemical groups used in a hydraulic fracturing job, to analyze its impact independently.
In order to understand the impact of hydraulic fracturing chemicals on short term (6 month BOE), midterm (12 month BOE) and long term (24 months BOE) productivity, an analysis is performed using feature selection methods like SelectKBest (F Regression score), Tree-based regression, Mutual Info Regression, Recursive Feature Elimination (RFE) and Correlation-based Feature Selection (CFS). In this paper, Max Ingredient Mass (LBS) is utilized for correlation with production to determine if more or fewer quantities should be used in the Powder River Basin. Comprehensive data exploration is carried out to generate correlations between the chemicals and productivity which helps to identify and link multiple causes to effects and weigh the strength of the correlations, which might not be obvious to the naked eye. A comparison of algorithms for three different target variables shows how certain chemicals are significant at the beginning of production and gain/lose their importance with time.
The present paper provides in-depth insight into the impact of the most commonly used chemical types in shale completions. It uses machine learning to provide a truly novel understanding of the individual contribution of these products. It adds to the understanding around this concept by developing a cause and effect relationship between chemical design and composition and the impact on production and provides recommendations on process optimization in completion design.