| 基于卷积神经网络(CNN)的泥质烃源岩TOC预测模型——以鄂尔多斯盆地杭锦旗地区为例 |
| |
| 作者姓名： | 王惠君 赵桂萍 李良 张威 齐荣 刘珺 |
| |
| 作者单位： | 1.中国科学院大学地球与行星科学学院, 北京 100049;2.中国科学院计算地球动力学重点实验室, 北京 100049;3.中国石油化工有限公司华北分公司, 郑州 450006;4.中石化华北石油工程有限公司测井分公司, 郑州 450006 |
| |
| 基金项目： | 中国石油化工股份有限公司华北分公司源项目（Y34101X1G2）资助 |
| |
| 摘 要： | 总有机碳含量（TOC）是烃源岩评价的重要指标。传统的TOC预测模型有ΔlogR和BP神经网络，但是ΔlogR的拟合精度较低，BP神经网络容易陷入局部最优。针对这些问题，提出一种基于卷积神经网络（convolutional neural network，CNN）预测烃源岩TOC的方法。以鄂尔多斯盆地杭锦旗地区上古生界泥质烃源岩为研究对象，通过对比实验验证该方法的有效性。实验结果表明，CNN可用于TOC预测，且预测精度高于ΔlogR和BP神经网络。利用CNN对108口钻井的山1段和太原组泥岩的TOC值进行预测，并结合沉积微相做出TOC平面图，发现研究区的东南部和中部的沼泽沉积微相的TOC值较高，分流河道沉积微相的TOC值较低。TOC值的平面分布与沉积微相分布在整体上具有良好的匹配关系，显示了CNN方法计算TOC的可行性。
|
| 关 键 词： | 泥质烃源岩 TOC 测井 卷积神经网络(CNN) BP神经网络(BPNN) |
| 收稿时间： | 2018-07-23 |
| 修稿时间： | 2018-11-29 |
TOC prediction model for muddy source rocks based on convolutional neural network (CNN): a case study of the Hangjinqi area of the Ordos Basin |
| |
| Authors: | WANG Huijun ZHAO Guiping LI Liang ZHANG Wei QI Rong LIU Jun |
| |
| Institution: | 1.College of Earth and Planetary Science, University of Chinese Academy of Sciences, Beijing 100049, China;2.Key Laboratory of Computational Geodynamics of Chinese Academy of Sciences, Beijing 100049, China;3.Geoscience Research Institute of North China Company, SINOPEC, Zhengzhou 450006, China;4.Well Logging Company of North China Petroleum Engineering Company, SINOPEC, Zhengzhou 450006, China |
| |
| Abstract: | Total organic carbon (TOC) is a significant factor for oil and gas exploration and development. For determining TOC, previous researchers mainly employed ΔlogR and back propagation artificial neural network (BP). However, ΔlogR has a low coefficient of the determination of the calculated TOC compared to the core measured data, while BP is easily trapped into a local optimum. To address these challenges, we propose a new method based on convolutional neural network (CNN) to calculate TOC. In this work, the argillaceous source rocks in the Hangjinqi area of the Ordos Basin were studied, and the effectiveness of the method was verified. The experimental verification showed that CNN could be used for TOC prediction of the source rocks and the prediction accuracy was higher than those of ΔlogR and BP neural network. CNN was used to predict the TOC values of the Shan 1 and Taiyuan mudstones of 108 wells. The TOC planar graphs were made in combination with the sedimentary microfacies. We found that the TOC values of the swamps in the southeast and central parts were higher and the TOC values of the distributary channels were generally lower. The TOC plane distribution well matched the sedimentary microfacies distribution, showing the feasibility of this prediction method. |
| |
| Keywords: | muddy source rock TOC content well logging convolutional neural network(CNN) BP neural network(BPNN) |
|
| 点击此处可从《》浏览原始摘要信息 |
| 点击此处可从《》下载免费的PDF全文 |
