Application of Interpretable Machine-Learning Workflows To Identify Brittle, Fracturable, and Producible Rock in Horizontal Wells Using Surface Drilling Data

Abstract

Summary In this study, we demonstrate the application of an interpretable (or explainable) machine-learning workflow using surface drilling data to identify fracturable, brittle, and productive rock intervals along horizontal laterals in the Marcellus Shale. The results are supported by a thorough model-agnostic interpretation of the input/output relationships to make the model explainable to users. The methodology described here can easily be generalized to real-time processing of surface drilling data for optimal landing of laterals, placing of fracture stages, optimizing production, and minimizing fracture hits. In practice, this information is rarely available in real time and requires tedious and time-consuming processing of logs (including image logs), core, microseismic data, and fiber-optic-sensor data to provide post-job validation of fracture and well placement. Post-completion analyses are generally too late for corrective action, leading to wells with a low probability of success and increasing risk of fracture hits. Our workflow involves identifying geomechanical facies from core and well-log data. We verify that the geomechanical facies derived using core and well-log data have characteristically different brittleness, fracturability, and production characteristics. We test and investigate several different supervised classifiers to relate surface drilling data to the geomechanical facies. The data were divided into training and test data sets, with supervised classification techniques being able to accurately predict the geomechanical facies with 75% accuracy on the test data set. The clusters predicted on test well (unseen data) were qualitatively verified using the microseismic interpretation. The use of Shapley additive explanations (SHAP) helps explain the predictive models, rank the importance of various inputs in the prediction of the facies, and provides both local and global sensitivities. Our study demonstrates that pre-existing natural-fracture networks control both the hydraulic-fracture geometry as well as the production. Natural fractures promote the formation of complex fracture networks with shorter half-lengths, which increase well productivity while minimizing fracture hits and neighboring-well interactions. The natural-fracture network is itself controlled by the geomechanical properties of the rock. The ability of the surface drilling data to reliably predict the geomechanical rock facies provides a powerful tool for real-time optimization of wellbore trajectory and completions.