[Objective] Aiming at the challenges of strong heterogeneity, complex lithology, and insufficient accuracy of traditional evaluation methods in the Qingshankou Formation shale reservoirs of the Songliao Basin. [Methods]This study proposes a multi-parameter collaborative prediction framework based on well-log data. By integrating an improved ΔlogR method, a Fully Connected Neural Network (FCNN), and optimized empirical formulas, efficient prediction models for Total Organic Carbon (TOC), mineral content, and porosity were established. The enhanced ΔlogR method addresses nonlinear mapping in TOC prediction for high-maturity shale through stratum-specific baseline calibration and dynamic adjustment of optimization coefficients. The FCNN model, utilizing six well-log parameters (including acoustic travel time and gamma ray), establishes a nonlinear inversion model for predicting siliciclastic, clay, and carbonate mineral contents. Porosity prediction was refined by calibrating core data to optimize a synergistic acoustic-density-neutron log calculation formula. [Results] Application examples demonstrate significant improvements: the improved ΔlogR method enhances TOC prediction accuracy, the mineral content model achieves an R2 of 0.77, and porosity calculations align well with core measurements. Innovatively combining geological prior knowledge with machine learning algorithms, this study develops an integrated parameter prediction system suitable for small-sample, complex shale reservoirs. [Conclusion] The framework provides theoretical and methodological support for comprehensive evaluation and efficient development of shale oil reservoirs in the Songliao Basin, offering a practical solution for low-data-density unconventional resource assessment.
DownLoad: