LAUSR

Shale reservoirs are characterized by low porosity and strong anisotropy. Conventional geophysical methods are far from perfect when it comes to the prediction of shale sweet spot locations, and even less reliable when attempting to delineate unconventional features of shale oil and gas. Based on some mathematical algorithms such as fuzzy mathematics, machine learning and multiple regression analysis, an effective workflow is proposed to allow intelligent prediction of sweet spots and comprehensive quantitative characterization of shale oil and gas reservoirs. This workflow can effectively combine multi-scale and multi-disciplinary data such as geology, well drilling, logging and seismic data. Following the maximum subordination and attribute optimization principle, we establish a machine learning model by adopting the support vector machine method to arrive at multi-attribute prediction of reservoir sweet spot location. Additionally, multiple regression analysis technology is applied to quantitatively predict a number of sweet spot attributes. The practical application of these methods to areas of interest shows high accuracy of sweet spot prediction, indicating that it is a good approach for describing the distribution of high-quality regions within shale reservoirs. Based on these sweet spot attributes, quantitative characterization of unconventional reservoirs can provide a reliable evaluation of shale reservoir potential.