Lithological classification for gas hydrate reservoirs in permafrost areas based on ensemble learning

Objective The accurate identification of stratigraphic lithology plays a critical role in delineating gas hydrate−bearing layers. This study applies ensemble learning methods to establish correlations between logging data and gas hydrate stratigraphic lithology, exploring techniques for precise prediction of stratigraphic lithology in gas hydrate−bearing layers within permafrost regions. Conventional well−logging lithology identification methods suffer from limitations such as heavy reliance on expert knowledge, high subjectivity, and poor reproducibility. Moreover, these traditional approaches lack adaptability, often requiring the reconstruction of interpretation models when applied to new regions, which impedes timely application.

Methods This study focuses on gas hydrate reservoirs in the permafrost regions of the Qinghai−Xizang Plateau, where an ADASYN−XGBoost ensemble learning model is developed for lithology identification. The performance of this model is compared with that of several non−optimized machine learning algorithms, including Extreme Gradient Boosting (XGBoost), Random Forest (RF), K−Nearest Neighbors (KNN), Gradient Boosting Decision Tree (GBDT), and Support Vector Machine (SVM).

Results The results indicate that the ADASYN−XGBoost model achieves the highest lithology identification accuracy of 97.8% for gas hydrate formations in permafrost regions, significantly surpassing the accuracy rates obtained by the XGBoost, RF, KNN, GBDT, and SVM models.

Conclusions The ensemble learning−based lithology identification model proposed in this study offers a theoretical foundation and technical support for addressing the challenges associated with lithological classification of gas hydrate reservoirs in permafrost regions.