快速准确地识别覆盖区下伏地层与岩体，对于金属矿山地质找矿工作具有重要意义。针对矿床地层与岩体中复杂岩性分布的多样性和非均衡性，考虑测井响应特征与岩性之间的强非线性关系，提出了一种基于ADASYN非均衡数据处理和CatBoost机器学习的测井岩性智能识别方法。首先，利用ADASYN算法处理非均衡测井样本数据，根据小类样本加权分布生成合成样本；然后，采用CatBoost算法结合网格搜索以及十折交叉验证建立最优岩性识别模型；最后，通过模型输出的特征重要性及部分依赖图对岩性分类结果进行解译。以胶西北招贤金矿床实例测井数据为基础，针对10类岩性进行识别和解译分析，模型评价结果表明：测试集上的精确率、召回率和F1分数分别达到98.21%、98.20%和98.20%。将CatBoost岩性分类与GBDT、LightGBM算法进行对比，结果表明CatBoost分类效果最优，且均优于样本数据未均衡化处理的岩性识别效果。通过与实例录井剖面岩芯岩性进行对比，验证了模型分类结果的有效性。

Logging lithology identification is helpful to quickly and accurately identify the underlying strata and rock mass in the overburden area，which is of great significance to the geological prospecting exploration of metal mines. Based on the actual logging data of the Zhaoxian gold deposit in the northwest of Jiaodong Peninsula，this paper combined machine learning methods to research on intelligent identification of lithology. In view of the diversity and non-equilibrium of lithology distribution of complex rock formations in the deposit，considering the strong non-linear relationship between logging response and lithology，this paper proposed an intelligent identification method for logging lithology based on ADASYN imbalanced data processing and CatBoost machine learning.Firstly，the ADASYN algorithm was used to process the unbalanced logging sample data and generate synthetic samples according to the weighted distribution of small class samples. Then，the CatBoost algorithm was used to construct a machine learning model between logging characteristic and lithology. The validation curve was used to determine the hyperparametric grid search range of the model. Parameters were optimized by combining grid search with grid search and 10-fold cross validation to establish the optimal lithology classification model.Finally，the performance of the model was evaluated by indices such as accuracy，recall and F1 score on the test set，while the results of the lithology classification were interpreted by the model output of the feature importance and the partial dependence map.An example was given on the logging data from the Zhaoxian gold deposit in northwest Jiaodong peninsula，the lithology identification and interpretation analysis were conducted on 10 types of lithologies based on sample data equalisation. The model evaluation results show that the accuracy，recall and F1 score on the test set reached 98.21%，98.20% and 98.20%，respectively.CatBoost lithology classification was compared with GBDT and LightGBM algorithms，and the results show that CatBoost classifier has the best performance and is superior to the lithology recognition effect of sample data without equalization processing.The comparison with the lithology of example logging section cores verifies the validity of the model classification results.The results of the feature importance of the model output indicate that the logging features contribute to lithology classification are resistivity，natural potential and natural gamma.The strong correlation between these logging features and the identification of the lithology is a good indication of further mineralization.

http://www.goldsci.ac.cn/article/2023/1005-2518/1005-2518-2023-31-5-721.shtml