An indirect method of fracturing pressure interpretation based on data-driven workflows

Abstract

Fracturing pressure is one of the most important measurements recovered during hydraulic fracturing operations and carries important information regarding proppant transport and formation properties. We propose an indirect pressure-interpretation method to extract useful information, in which fracturing pressure is conversely used as output, rather than as input in conventional methods. A data-driven workflow is established to predict fracturing pressure by combing three methods (numerical models, machine learning algorithms, and error analysis) based on a dual-element hypothesis (injection and formation elements that control pressure variations). The numerical models extract injection and formation features for training the machine learning algorithms, then are promoted or optimized by analyzing their contribution in reducing prediction errors. The advanced models, in consequence, boost numerical characterizations of injection safety and formation evaluation. Two cases are presented to illustrate applications of the new method in the evaluation of proppant injection and brittleness index. A new mechanism of proppant transport and new brittleness calculation are reported by processing field data. The new method, integrating different numerical models, is compatible with broader applications in the evaluation of engineering and geological responses to hydraulic fracturing.