The accurate estimation of the elastic properties of the rock is of great importance for designing a successful hydraulic fracturing. Among these properties, Young's modulus and Poisson's ratio essentially control fracture aperture and conductivity. However, the fissile nature of the shale rock largely challenges the mechanical properties measurement using a cylindrical core sample. While the nanoindentation technology can be applied to measure small chips of rock fragment, but reproducible experiments are required to provide an unbiased estimation. Herein, we are proposing a machine learning approach to predict the elastic moduli. We utilized an ensemble of data mining techniques and a database that include both the mineralogy and pore characteristics. Our results indicate that K-Means clustering yields best performance on data classification than all other tested methods while the elastic moduli estimation from Artificial Neural Network (ANN)is most accurate than Support Vector Machine (SVM), Multivariate Linear Regression (MLR) and Multivariate Adaptive Regression Spine (MARS). The dimension reduction became essential when then input datasets are remarkably correlated. The supervised learning techniques with our proposed approach leverage the usability of the lab experiment data and overcome disadvantages of the traditional elastic moduli measurement. It also further lands the far-reaching guide for the fracturing design.


Machine learning have recently revolutionized the oil and gas industry (Alcocer and Rodrigues 2001, Al-Fattah and Startzman 2001, Kohli and Arora 2014, Okpo et al. 2016, Sinha et al. 2016, Tariq et al. 2017, Luo et al. 2018, Nande 2018, Rashidi et al. 2018, Sidaoui et al. 2018, Xu et al. 2019). As a data-rich industry, machine learning finds applications in every corner ranging from production forecast to drilling efficiency (Hegde and Gray 2017, Fulford et al. 2016). Given the significance of geomechanical properties of the rock, the volume of studies has attempted to leverage machine learning techniques. For instance, Li et al. (2018) developed a workflow implementing various machine learning algorithm to accurately provide an alternative to synthesize the sonic logs and geomechanical properties afterwards. In the same time, Hadi and Nygaard (2018) used Artificial Neural Network (ANN) to develop an empirical model to estimate the shear velocity from conventional logs. Another dimension was presented by Jain et al. (2015) where they proposed an approach to integrate both core and log spectroscopy which provided better estimations of the mineralogy.

This content is only available via PDF.
You can access this article if you purchase or spend a download.