Reservoir characterization by X-ray fluorescence（XRF）analysis and machine learning in tight reservoir

Abstract

Development of tight reservoir is represented by multi-stage hydraulic fracturing and horizontal drilling. Optimized hydraulic fracturing design based on the geomechanical character of horizontal section is required in order to achieve efficient oil and gas recovery from tight reservoir. Generally, geomechanical analysis using core sample and sonic logging are helpful to capture a geomechanical feature, however, application of these technique is too costly way to obtain an enormous quantity of data from many producing wells. The use of X-ray fluorescence（XRF）-related techniques in shale plays has been standard method to evaluate rock character of fine-grained rocks due to the difficulties in visual characterization. For example, in the Eagle Ford play, there are many XRF-related studies, such as estimation of depositional environments, mineral composition, total organic carbon（TOC）, etc. In this study, we constructed the prediction model of mineral composition, TOC and geomechanical features using machine learning methods based on XRF data obtained from conventional core of a vertical pilot well in Eagle Ford. The constructed prediction model was then applied to the XRF data of horizontal well’s cuttings samples. Multi-regression, random forest and auto machine learning（Auto ML）were utilized to estimate mineral composition, TOC and geomechanical character in this study. Modified multi-regression model for estimation of mineral composition and TOC achieved similar quality with recent researches, and random forest and Auto ML model（tree algorism）were able to estimate Leeb hardness, Young’s modulus and Poisson’s ratio with high accuracy. As the workflow constructed with Eagle Ford sample worked well in Onnagawa Formation, this prediction techniques using machine learning methods is expected to be applicable for various tight reservoir plays.