多元线性回归模型与多层感知器神经网络在铀矿测井泥质含量预测中的应用  

Application of Multiple Linear Regression and Multilayer Perceptron Neural Network in Predicting Shale Content for Uranium Logging

作　　者：张喆安 刘龙成 王书黎 白云龙 谢廷婷[1] ZHANG Zhean;LIU Longcheng;WANG Shuli;BAI Yunlong;XIE Tingting(Beijing Research Insititute of Chemical Engineering and Metallurgy,CNNC,Beijing 101149,China;Hohai University,Nanjing,Jiangsu 21000,China)

机构地区：[1]核工业北京化工冶金研究院,北京101149 [2]河海大学,江苏南京210000

出　　处：《铀矿地质》2024年第5期1007-1013,共7页Uranium Geology

基　　金：中核集团基础研究项目“电驱动强化地浸机理研究”(编号:CNNC-JCYJ-202333)资助。

摘　　要：在铀矿资源勘探工作中,泥质含量的测定对于确定地下岩层的性质和砂岩型铀矿床的分布具有重要意义。文章旨在避免常规测井解释计算方法受到希尔奇系数选取准确性的限制,提出了利用多元线性回归模型和多层感知器(MLP,Multilayer Perceptron)神经网络对测井数据进行分析与预测的方法。通过选取某地区的测井数据,采用多元线性回归模型和MLP神经网络进行了泥质含量关系模型的构建和验证。结果显示,多元线性回归模型在泥质含量低层位出现过拟合现象,而MLP神经网络则表现出更高的预测准确性,MLP神经网络在泥质含量预测中优于传统多元线性回归模型,为铀矿勘探中泥质含量的准确预测提供了有效工具,并有望改进现有的泥质含量评价方法。这些研究成果可显著提升测井解释的效率和准确性,对后续铀矿勘探开发工作的开展具有积极影响。The determination of shale content in uranium exploration is crucial for identifying the properties of underground rock formations and the distribution of sandstone-type uranium deposits.This study aims to overcome the limitations of traditional logging interpretation methods,which are affected by the accuracy of the selected GCUR coefficient,by proposing the use of multiple linear regression(MLR)models and multilayer perceptron(MLP)neural networks for analyzing and predicting logging data.By selecting logging data from a specific region,we constructed and validated shale content relationship models using MLR and MLP neural networks.The results indicate that the MLR model exhibits overfitting in low shale content intervals,whereas the MLP neural network demonstrates higher prediction accuracy.The MLP neural network outperforms the traditional MLR model in predicting shale content,providing an effective tool for accurate shale content prediction in uranium exploration.This approach promises to improve existing shale content evaluation methods significantly,enhancing the efficiency and accuracy of logging interpretation,and positively impacting subsequent uranium exploration and development activities.

关 键 词：铀矿测井 泥质含量 多元线性回归模型 多层感知器神经网络 

分 类 号：P631.84[天文地球—地质矿产勘探]