An integrated machine learning-based approach to identifying controlling factors of unconventional shale productivity

The controlling factors of unconventional shale productivity by comprehensive analysis of mineralogy, petro-physics, geochemistry, and geomechanics have not been well understood. The comprehensive datasets from 1182 core samples of key wells from the Duvernay shale at Crooked Lake, Alberta, are gathered to evaluate the fundamental parameters controlling unconventional shale gas production. By integrating reservoir parameters and shale productivity, a machine learning-based approach is used to identify the fundamental elements that affect shale productivity. Four machine learning approaches are evaluated, where Extra Trees has led to the highest coefficient of determination R2 of 0.817. Factors that mostly contribute to shale productivity are found to be the production index, formation pressure, effective porosity, total organic carbon, gas saturation, and shale thickness. Case studies demonstrate that the average accordance rate between the predicted and actual pro-duction of three new wells reaches 92.3%, thereby shedding light on the site selection of hydraulic fracturing wells for the efficient development of unconventional resources.

Automated summary

The controlling factors of unconventional shale productivity have not been well understood and this study aims to evaluate these factors using comprehensive datasets from 1182 core samples of key wells from the Duvernay shale in Alberta. By integrating reservoir parameters and shale productivity, a machine learning-based approach is used to identify the fundamental elements affecting shale productivity. The results show that factors such as production index, formation pressure, effective porosity, total organic carbon, gas saturation, and shale thickness contribute significantly to shale productivity.