Investigation of hydraulic fracture propagation in conglomerate rock using discrete element method and explainable machine learning framework.

Conglomerate reservoirs are important oil and gas resources that require hydraulic fracturing for stimulation. However, the heterogeneity of conglomerate rocks causes fractures to propagate irregularly, complicating the fracturing design. To provide insight into the complex fracture behavior in conglomerate rocks, a three-dimensional (3D) hydromechanical numerical model based on the discrete element method (DEM) was proposed in this study. Besides, a novel approach combining the grain-based DEM with Voronoi tessellation was adopted to depict the geometrical characteristics of conglomerate rocks. Considering the rock matrix-interface-gravel structure, the effects of various influencing factors including the strength and permeability, in situ stress difference, fluid properties and injection scheme on the fracture propagation and induced microseismic events were investigated. Two fracture behaviors, penetration and deflection, were summarized. Finally, the mechanisms of different fracture behaviors were discussed, revealing the joint effects of various factors on the fracture behavior. To predict the fracture behavior in conglomerate rocks, an explainable machine learning framework comprising the extreme gradient boosting and the shapley additive explanations was adopted, which attained high accuracy on the testing set. It can also provide comprehensive explanations for the predicted results, offering support for practical decisions. [ABSTRACT FROM AUTHOR]