| 深度学习在煤层气测井解释中的应用研究 |
| |
| 引用本文： | 向,旻.深度学习在煤层气测井解释中的应用研究[J].地质与勘探,2020,56(6):1305-1312. |
| |
| 作者姓名： | 向 旻 |
| |
| 作者单位： | 新疆工程学院矿业工程与地质学院，新疆乌鲁木齐 |
| |
| 基金项目： | 新疆维吾尔自治区高校科研计划项目（编号：XJEDU2017S056）、新疆维吾尔自治区自然科学基金项目 (编号：2017D01B08)、新疆维吾尔自治区“百名青年博士引进计划”项目（编号：BS2017001）资助 |
| |
| 摘 要： | 将常规储层测井解释方法应用于煤层气储层测井解释，其效果存在一定的折扣。为了改善传统方法在煤层气测井解释中出现的问题，将深度学习的思想引入测井解释，提出受限玻尔兹曼机的数量、受限玻尔兹曼机隐含层神经元数量、分类阈值的确定方法，利用深度信念网络进行煤层识别及煤层气含气量的预测。实验结果表明：首先，在交会图法效果不好的情况下，通过深度信念网络进行煤层识别，继而对识别结果进行适当校正，煤层识别成功率可达到90%以上；其次，经过多种方法的对比，利用深度信念网络进行煤层气含气量预测的效果，要好于BP神经网络、多元回归统计以及Langmuir方程三种方法。深度学习改进了传统的BP神经网络，具备更强的复杂函数泛化能力，适用于煤层气测井解释，并具有进一步的推广价值。
|
| 关 键 词： | 深度学习 煤层气测井 煤层识别 含气量 |
| 收稿时间： | 2019/7/21 0:00:00 |
| 修稿时间： | 2020/8/3 0:00:00 |
Application of deep learning to coalbed methane logging interpretation |
| |
| Xiang Min.Application of deep learning to coalbed methane logging interpretation[J].Geology and Prospecting,2020,56(6):1305-1312. |
| |
| Authors: | Xiang Min |
| |
| Institution: | College of Mining Engineering and Geology, Xinjiang Institute of Engineering, Urumqi, Xinjiang |
| |
| Abstract: | The effect will be discounted when applying conventional reservoir logging interpretation to coalbed methane reservoirs. In order to solve this problem, the deep belief network, one method of deep learning, is introduced into coalbed methane logging interpretation, which is used to identify coal seams and predict coalbed methane content. Then this paper proposes the methods to determine the quantities of restricted Boltzmann machine and the neurons amount in the hidden layer of restricted Boltzmann machine and the threshold of lithology classification to train the deep belief network. Experimental results show that, firstly, in the case of poor effect of the cross plot, coal seam can be identified by the deep belief network with the reasonable correction, which can result in a successful rate of identification more than 90%. Secondly, through the comparison of various methods, the effect of using deep belief network to predict coalbed methane content is better than that of BP neural network, multivariate regression statistics and Langmuir equation. Therefore, deep learning improves the traditional BP neural network and has stronger generalization ability of complex functions, which is suitable for coalbed methane logging interpretation and has a value of further population. |
| |
| Keywords: | deep learning coalbed methane logging coal seam identification coalbed methane content |
|
| 点击此处可从《地质与勘探》浏览原始摘要信息 |
| 点击此处可从《地质与勘探》下载免费的PDF全文 |
|
