یک روش جدید برای ارزیابی رطوبت سنگ در تشکیل نفت تنگ از داده های معمولی و داده های پتروفیزیکی

Brittleness is a critical indicator for hydraulic fracturing candidate screening in unconventional reservoirs. Current rock brittleness estimation models are often inferred from mechanical parameters and mineralogical data, which primarily use empirical equations. However, the absence of shear sonic velocity data and insufficient mineral data sometimes restricts its wide application. In this article, our objective is to illustrate the application of a data-driven approach for rock brittleness estimation that employs computational intelligence technologies (multilayer perception and radial basis function models) that use conventional well logs as inputs. To reflect the local rock type variation with depth, we first updated the typical mineralogy based brittleness calculation formulas. A database of the well logs, mechanical parameters, X-ray diffraction (XRD) and QEMSCAN mineralogy results collected from a single well in the Santanghu tight oil formation in the Xinjiang basin, China was then constructed. Rock brittleness tests were performed using a multilayer perception model and radial basis function model with different inputs. The comparison of the rock brittleness results produced by the log-based soft computing technologies, mechanical-based method and mineralogy-based method demonstrated that the data-driven approach is flexible and has sufficient accuracy. According to the performance indicators, the predictive performance of the radial basis function model was found to be better than that of the multilayer perception model. This study shows that soft computing technologies can be used to infer missing data when the mineralogical data are inadequate and are less dependent on acoustic full-wave logging, and they are therefore more applicable and practical than traditional empirical formulas.