Detecting thermal water layer with algorithm model utilizing well-logging reconstruction data: A case study of the Qingshankou formation, Songliao Basin

Abstract

The conversion of “oil-geothermal” has become a hot topic in regions where petroleum reserves have been depleted but possess moderate to high ground heat flow values. This shift necessitates a rapid and objective redefinition of thermal water layers (WL), which were often overlooked during the oil exploration stage. In this study, we aimed to establish a rapid WL identification model based on traditional well-logging data, corrected for external influencing factors. Our findings suggest that: (1) optimizing well-logging data before developing algorithmic models is crucial, as it enhances the outcomes; (2) reconstructing well-logging data significantly increases the accuracy of algorithmic models from 67.24% to 92.41%, after accounting for the effects of clay minerals, cementation, drilling mud invasion, and variations in temperature and salinity; and (3) integrating different algorithmic models also improves identification accuracy (92.41% increased to 92.76%) and reduces the misidentification of oil layers (OL) as WL compared to single models (13.6% reduced to 4.5%). However, the order in which these algorithms are applied is important. Notably, the significant overlap in resistivity of the CaCl₂ type, primarily between 0.6 to 0.8 Ω·m, contributes to the highest rate of misidentifying OL as WL. And the new well also validated the universality of our models, demonstrating a high precision in WL identification and a low rate of misidentifying OL as WL. Although the established model enables rapid and objective WL identification, there are still some deficiencies that need to be addressed, especially in lithofacies character, layer thicknes and newest algorithm model application.