Intelligent Identification Method for Deep Carbonate Rock Well Logging Lithofacies Incorporating Sequence Stratigraphic Prior Information

Lithofacies identification is the core component of reservoir characterization and holds significant importance for oil and gas exploration and development. Traditional lithofacies identification primarily relies on expert experience and manual interpretation, which suffers from strong subjectivity, low efficiency, and poor consistency, making it difficult to meet the demands of refined development in actual oil and gas fields. This paper proposes an intelligent logging lithofacies identification method for deep carbonate rocks that integrates sequence stratigraphic prior information. Research demonstrates that, compared to manual interpretation methods, deep learning-based intelligent lithofacies classification technology can significantly enhance lithofacies identification efficiency. Using a standard consumer-grade GPU, the inference task for lithofacies identification across over forty wells can be completed in approximately four minutes. Compared to deep learning approaches that do not incorporate sequence prior information, the new method achieves a test accuracy improvement of around 35%. The proposed technique has yielded promising results in the carbonate lithofacies classification task of the Changxing Formation in the Yuanba gas reservoir, with a blind well test accuracy nearing 95%. The intelligent logging lithofacies classification method presented in this paper has reached considerable practicality and can effectively improve the efficiency of oil and gas exploration and development.