In complex sedimentary environments, the identification of thin sandbodies and the accurate prediction of their thickness remain challenging, particularly when relying on a single analytical approach. Taking the lower sub-member of the fourth member of the Shahejie Formation (Es4L) in the Chenguanzhuang area of the Dongying Depression as a case study, this study proposes a quantitative prediction method that integrates sedimentary facies constraints with machine learning-based seismic multi-attribute fusion. Based on core observations, well log data, and 3D seismic datasets, the study area is subdivided into two zones: Zone I (shallow-water delta front) and Zone II (shore–shallow lake). Sensitive attributes for each zone are optimized using Pearson correlation analysis and hierarchical clustering, and five machine learning models—SVR, Random Forest, MLP, Ridge Regression, and Lasso Regression—are systematically evaluated. The MLP model is selected for Zone I, achieving R2 values of 0.856 and 0.936 for the training and test sets, respectively, whereas Ridge Regression combined with leave-one-out cross-validation (LOOCV) is adopted for Zone II to mitigate overfitting caused by limited well data, yielding R2 values of 0.864 and 0.779. Compared with conventional linear regression (R2 = 0.45), the proposed approach significantly improves the accuracy of quantitative sandbody prediction, providing a reliable geological basis for hydrocarbon exploration and an effective technical framework for similar complex sedimentary environments.
Liu et al. (Mon,) studied this question.