Deep marine shale reservoirs are controlled by multi-factor coupling effects, and the genetic mechanism of “sweet spots” exhibits strong complexity, leading to prominent difficulties in quantitative prediction and precise evaluation of sweet spots. Aiming at the problems of an unclear lithofacies-controlled sweet spot evolution law and insufficient accuracy of multi-parameter quantitative evaluation in traditional evaluation methods, this paper takes the Wufeng Formation and Long1 member of the Longmaxi Formation in the LZ block, Southern Sichuan, as the research object. Innovatively integrating machine learning (ML), grey correlation analysis (GRA), and three-dimensiona (3D) geological modeling technologies, a refined prediction model for reservoir sweet spot evaluation indicators under lithofacies constraint conditions is established, and a multi-parameter fusion quantitative evaluation method for deep marine shale gas sweet spots with high prediction accuracy is proposed. The results demonstrate that the LightGBM-based prediction model for sweet spot evaluation indicators achieved excellent performance. Based on a total of 380 preprocessed samples divided into training and test sets in a 7:3 ratio, the coefficient of determination (R2) of the model exceeded 0.9 in both the test and validation datasets. The “sweetness index”, a comprehensive evaluation index of reservoir sweet spots constructed via GRA-based multi-factor fusion, shows a correlation coefficient of 0.91 with respect to actual gas well production, presenting a high fitting degree. The 3D sweet spot geological model reveals that Class I sweet spots are mainly developed in the 1st to 3rd sub-layers of the Long1 member, while Class II sweet spots are distributed in the 5th and 6th sub-layers, which is highly consistent with the actual development law of the gas field. This study breaks through the limitations of single evaluation methods and weak lithofacies control consideration in traditional sweet spot evaluation and forms a set of innovative technical process integrating “precision prediction—multi-factor fusion—3D characterization”. It provides a new technical approach for efficient and accurate evaluation of deep marine shale reservoir sweet spots and has important guiding significance for the efficient development of shale gas.
Liu et al. (Sat,) studied this question.