Identifying lithology from well logs is an important step in deep prospecting and resource estimation. Various machine learning algorithms have been adopted to identify lithology in oil and gas fields. Such algorithms, however, are rarely used for mineral deposits because of their complex geological conditions. In this paper, we propose an application framework using the gradient boosting decision tree (GBDT) algorithm to identify lithology from well logs in a mineral deposit. The GBDT classifier was built via the procedure of grid search and cross-validation to optimize the hyperparameters. In the Zhaoxian gold deposit, as the study area, an optimized GBDT classifier was built to fit the association between a set of well logs and ten lithological classes. The results demonstrate that the GBDT classifier has good classification performance in lithology identification, with a precision of 93.55%, a recall of 93.49% and an F1-score of 93.27%. The GBDT classification results also indicate that the major features contributing to the lithology classification are resistivity, followed by spontaneous potential and natural gamma according to the model interpretation of feature importance and partial dependence plots. The study demonstrates that the GBDT model can enhance our understanding of lithology identification from well logs in mineral deposits, which provides significant implications for further exploration targeting the deep-seated parts of mineral deposits.

从测井中识别岩性是深部勘探和资源估算的重要步骤。已采用各种机器学习算法来识别油气田中的岩性。然而，由于其复杂的地质条件，此类算法很少用于矿床。在本文中，我们提出了一个应用框架，使用梯度提升决策树 (GBDT) 算法从矿床的测井中识别岩性。GBDT 分类器是通过网格搜索和交叉验证的过程构建的，以优化超参数。以昭县金矿为研究区，建立了优化的GBDT分类器，以拟合一组测井与10个岩性类别之间的关联。结果表明，GBDT分类器在岩性识别中具有良好的分类性能，准确率为93.55%，召回率为93.49%，F1-score为93.27%。GBDT 分类结果还表明，根据特征重要性和部分依赖图的模型解释，对岩性分类有贡献的主要特征是电阻率，其次是自发势和自然伽马。该研究表明，GBDT 模型可以增强我们对矿床测井岩性识别的理解，这对于进一步针对矿床深部进行勘探具有重要意义。GBDT 分类结果还表明，根据特征重要性和部分依赖图的模型解释，对岩性分类有贡献的主要特征是电阻率，其次是自发势和自然伽马。该研究表明，GBDT 模型可以增强我们对矿床测井岩性识别的理解，这对于进一步针对矿床深部进行勘探具有重要意义。GBDT 分类结果还表明，根据特征重要性和部分依赖图的模型解释，对岩性分类有贡献的主要特征是电阻率，其次是自发势和自然伽马。该研究表明，GBDT 模型可以增强我们对矿床测井岩性识别的理解，这对于进一步针对矿床深部进行勘探具有重要意义。