关键词: 岩性识别')">

岩性识别, 机器学习, 随机森林, 极端随机树, 平衡数据

Abstract: In the domains of oil and gas exploration and geoengineering, precise lithology identification holds paramount importance for the assessment and utilization of resources. The inherent complexity of geologic data and the imbalanced distribution of lithology samples pose significant challenges to traditional methods in terms of lithology identification. In this paper, we propose a methodology for lithology identification that combines SMOTE (synthetic minority over-sampling technique) with extra trees. Firstly, the SMOTE method is employed to enhance the representation of minority class samples, thereby improving the balance of the training data. Secondly, the lithology classification model is constructed using the high efficiency and strong generalization ability of extra trees. The experimental findings demonstrate that the recognition accuracy of extra trees is 85.54%, which is 5.58%, 2.55%, 2.35%, and 2.08% higher than that of other machine learning methods—gradient boosting decision tree (GBDT), extreme gXGBoost), light gradient boosting machine (LightGBM), and random forest method, respectively. The prediction bias of the model caused by sample imbalance is mitigated by SMOTE sampling, resulting in enhanced recognition accuracy for specific lithology categories within each model. Consequently, this leads to an overall enhancement in the performance of the model. The extra trees model exhibits the best performance, achieving an identification accuracy of 86.62%, which represents improvements of 4.71%, 2.56%, 1.55%, and 2.02% over GBDT, XGBoost, LightGBM, and random forest, respectively. These results confirm the effectiveness of combining SMOTE with extra trees for lithology identification.

Key words: lithology identification')">

lithology identification, machine learning, random forest, extra trees, data balancing