岩性识别是储层精细评价的基础，传统方法一般仅用2~3种测井参数的交互关系进行岩性识别，测井信息利用率低，对于岩性测井响应差异小的地层岩性识别精度低，严重制约了老井复查效果。高效的智能分类算法CatBoost可充分挖掘多源测井信息与岩性的关联。

以新疆轮南地区侏罗系砂泥岩储层为研究对象，通过敏感性分析选取自然伽马、自然电位、深浅电阻率比值、声波时差和密度5个测井参数，构建基于CatBoost算法的岩性智能识别模型。利用优化的模型处理实际井资料以进行地层岩性识别，通过准确率、精确率和召回率综合评估模型的岩性识别效果，并对比分析了其与随机森林和KNN算法模型的识别效果。

结果表明：轮南侏罗系大类岩性包括泥岩、砂岩和砾岩，细分岩性复杂；根据岩性敏感测井参数利用CatBoost算法建立的岩性智能预测模型对目标储层细分岩性的识别准确率达92.64%，显著高于随机森林模型的82.95%和KNN模型的70.16%，证明该方法能有效解决研究区的岩性识别问题。研究成果不仅为轮南地区老井复查和进一步勘探开发提供了科学依据，还为复杂岩性精细识别方法研究提供重要参考。

Lithology identification lays the foundation for fine-scale reservoir evaluation. However, traditional identification methods generally utilize the interactive relationships between only 2‒3 logging parameters, suffering from low utilization rates of logging information and low identification accuracy for strata with small differences in logging responses. This seriously restricts the effects of old well reexamination. The efficient, intelligent CatBoost classification algorithm can fully mine the correlations between multi-source logging information and lithology.

This study investigated the Jurassic sandstone and mudstone reservoirs in the Lunnan area, Xinjiang, China. Using five logging parameters determined through sensitivity analysis, i.e., natural gamma-ray value, spontaneous potential, deep and shallow resistivity ratio, sonic interval transit time, and density, this study developed an intelligent lithology identification model based on the CatBoost algorithm. The optimized model was employed to deal with actual logging data for lithology identification, and its performance was evaluated using accuracy, precision, and recall and was then compared with the lithology identification results of the random forest (RF) and k-nearest neighbors (KNN) algorithms.

The results indicate that the large-scale lithologies of the Jurassic strata in the Lunnan area include mudstones, sandstones, and conglomerates, with complex fine-scale lithologies. In the identification of the fine-scale lithologies of the target reservoir, the intelligent lithology identification model, established using the CatBoost algorithm and lithology-sensitive logging parameters, yielded an accuracy of 92.64%, significantly higher than that of the random forest model (82.95%) and the KNN model (70.16%). This result demonstrates that the CatBoost model can effectively address of the challenges of lithology identification in the study area. The results of this study will provide a scientific basis for the review and further exploration and development of old wells in the Lunnan area. Besides, these results can serve as a valuable reference for research on methods for fine-scale identification of complex lithologies.