Slickwater, Hybrid, Self-Suspending Proppant: Practical Use of Data Analytics to Determine Hydraulic Fracture Fluids

Abstract

Abstract Utilizing fluids that have friction reducing qualities and/or viscosity building properties are desirable for hydraulic fracturing in multi-stage horizontal wells. Horizontal wells targeting the Middle Bakken used various fluid types for stimulation were analyzed in this study. This paper uses data analytics to show that the answer to "which fluid type is better?". By applying statistical methods of k-means clustering, t-tests as well as multivariate analysis, a more robust answer was obtained in the comparison between slickwater, crosslinked gel, linear gel, hybrid systems, and Self-Suspending proppant. The first part of the analysis compared independent variables (lateral length, true vertical depth (TVD), liquid loading, sand loading, and fluid type) to well productivity in 532 Middle Bakken wells. Multivariate analysis was conducted to determine the dominant variables. Based on those, the data were filtered in two different ways: first on sand loading and secondly on lateral length and TVD. The second part of the analysis applied distinct k-means clustering. A multivariate model was built to accommodate the influence of each variable in each cluster. Having found inconclusive results, ANOVA (Analysis of Variance) analysis was conducted to analyze which fluid type resulted in the best overall well production. Several conclusions are demonstrated in this paper. First, industry stimulation treatment data are rarely crafted with thorough Design of Experiment rigor. This challenges many of the assumptions in statistics and data analytics to provide for unbiased analysis. Second, while lateral length, TVD and total sand are all seemingly independent variables physically, they were found to have a certain degree of statistical collinearity; a measure of variable dependence. Third, it was shown that wells using slickwater and 1Self-Suspending Proppant outperform wells using other proppant types in terms of overall production in the Middle Bakken. Hence, while the perfect frac job still eludes us, establishing a framework of unbiased analysis provide us with a robust approach to answer one of the key questions of our day: which fluid type is better? The novel thing about this type of analysis is that it takes thousands of different types of data points and narrows down what the most important variables in the data are. In an age where data is king, this approach follows statistic principles designed for handling such data.