Precise characterization of fluid‐sensitive logging responses in complex shale reservoirs is very important. It is essential for quantitative formation evaluation and development optimization. Based on a case study of the shale gas reservoir in the second member of the Wujiaping formation (hereinafter referred to as the Wu‐2 Member) in the Hongxing area of the Sichuan Basin, this paper proposes an integrated reservoir evaluation system combining parameter model optimization and machine learning. The study addresses the challenges of calculating reservoir parameters and identifying sweet spots exhibiting shale reservoirs characterized by high carbonate content, high total organic carbon (TOC), high gas content, anomalous resistivity, and thin interbedding. The workflow comprises two main stages. First, four key petrophysical models are enhanced to improve the accuracy of fluid and mechanical property predictions. The specific model is as follows: (1) A TOC calculation model based on a combination of ridge regression and the least squares method; (2) A gas saturation evaluation model based on the neutron‐density overlay method; (3) A calculation model of adsorbed gas corrected for temperature and Langmuir pressure; (4) A brittleness index calculation model based on the dual correction of carbonate mineral content and TOC. Second, precise reservoir classification and evaluation are achieved through a “Random Forest + Clustering” machine learning algorithm. Research indicates that the quantitative criteria for identifying “sweet spots” in the study area are established as follows: TOC >5%, gas saturation >50%, porosity >4.5%, and brittleness index >50%. Based on this standard, the average thickness of sweet spots is 16.03 m (accounting for 81.78% of the total). This framework provides a valuable reference for the exploration and development of analogous gas reservoirs.
Deng et al. (2026) studied this question.