基于2种机器学习方法的页岩TOC含量评价——以川南五峰组—龙马溪组为例

为了建立合理准确的川南五峰组—龙马溪组页岩TOC含量预测方法，以长宁、泸州等地区的测井曲线及17口井实测TOC含量数据为基础，利用主成分分析法对这些资料进行预处理，基于BP神经网络和梯度提升决策树（GBDT）方法建立2种TOC含量预测模型，并将之与传统TOC含量预测方法进行对比。结果表明： ① 2种新模型的准确度均高于传统方法，预测结果与实际值吻合度均满足要求。②与BP神经网络模型相比，GBDT预测精度更高，均方根误差仅为0.0387。利用GBDT方法所建立的TOC含量预测模型具有低成本、高效、连续等特点，能够快速准确地预测目的层TOC含量。该成果可为提高页岩油气勘探开发效率提供有效技术支撑。

Evaluation of shale TOC content based on two machine learning methods:A case study of Wufeng-Longmaxi Formation in southern Sichuan Basin

In order to establish a reasonable and accurate prediction method of shale total organic carbon(TOC) content of Wufeng-Longmaxi Formation in southern Sichuan Basin,the principal component analysis method was used to preprocess the logging curves and the measured TOC content data of 17 wells in Changning and Luzhou areas. Two TOC content prediction models were established based on BP neural network and gradient boosting decision tree(GBDT),and compared with the traditional TOC content prediction methods. The results show that:(1) The accuracy of the two models is higher than that of the traditional methods,and the consistence between the predicted results and the actual values can meet the requirements.(2) Compared with BP neural network model, GBDT has higher prediction accuracy,and the root mean square error is only 0.0387. The TOC content prediction model established by GBDT has the characteristics of low cost,high efficiency and continuity,and can be used to predict the TOC content of the target layer quickly and accurately. This achievement can provide effective technical support for improving the efficiency of shale oil and gas exploration and development.