| 采用支持向量回归从测井曲线定量计算油页岩含油率 |
| |
| 引用本文： | 陈敬武,朱建伟,孙平昌.采用支持向量回归从测井曲线定量计算油页岩含油率[J].地质与资源,2017,26(2):157. |
| |
| 作者姓名： | 陈敬武 朱建伟 孙平昌 |
| |
| 作者单位： | 1. 吉林大学 地球科学学院, 吉林 长春 130061;2. 油页岩与共生能源矿产吉林省重点实验室, 吉林 长春 130061;3. 东北亚生物演化与环境教育部重点实验室, 吉林 长春 130061 |
| |
| 基金项目： | 国家油气专项项目(第二批)，教育部产学研用合作创新项目 |
| |
| 摘 要： | 获取油页岩含油率是油页岩资源评价的重要步骤,传统从测井曲线计算油页岩含油率多采用回归模型,但存在误差大或过拟合的局限性和弱点.本文尝试结合大数据概念的数据挖掘算法和测井应用知识进行油页岩含油率定量计算,提高含油率计算的精度以及模型的泛化性.利用改进的ΔlogR技术获得DT_s、DEN_s、GR_s作为解释变量.采用数据挖掘算法——支持向量回归进行定量计算油页岩含油率能够大幅提高泛化性和精度,获得模型训练样本R~2得分为0.82,测试样本R~2得分可达0.70,拟合精度较高.支持向量回归模型比传统回归模型泛化能力更强,能够避免过拟合问题,具有广泛的应用性.
|
| 关 键 词： | 油页岩含油率计算 支持向量回归 改进的ΔlogR |
| 收稿时间： | 2017-01-13 |
QUANTIFYING OIL CONTENT OF OIL SHALE FROM WELL LOGS USING SUPPORT VECTOR REGRESSION |
| |
| CHEN Jing-wu,ZHU Jian-wei,SUN Ping-chang.QUANTIFYING OIL CONTENT OF OIL SHALE FROM WELL LOGS USING SUPPORT VECTOR REGRESSION[J].Geology and Resources,2017,26(2):157. |
| |
| Authors: | CHEN Jing-wu ZHU Jian-wei SUN Ping-chang |
| |
| Institution: | 1. College of Earth Science, Jilin University, Changchun 130061, China;2. Key Laboratory for Oil Shale and Coexistent Energy Minerals of Jilin Province, Changchun 130061, China;3. Key Lab for Evolution of Past Life and Environment in Northeast Asia, Ministry of Education, Changchun 130061, China |
| |
| Abstract: | The oil content is an important evaluation index for oil shale resources.Traditionally,calculation of the oil content from well logs of oil shale is performed by regression model,which,however,has the limitation and weakness of big error or over-fitting.This paper attempts to integrate the classical data mining algorithm with "Big Data" concept and logging application knowledge for oil content quantification to improve the accuracy and generalize the model.The explanatory variables of DTs,DENs and GRs for analysis are obtained by the improved △logR technique.The data mining algorithm of support vector regression (SVR) can greatly improve the model generalization and precision in the oil content quantification.The R2 score of training samples in the model is 0.82.A high fitting precision is achieved in the test samples,of which the R2 score is 0.70.The SVR model is more generalized than traditional regression model,and can avoid over-fitting problem and well applied. |
| |
| Keywords: | oil shale oil content quantification support vecter regression (SVR) improved △logR |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
| 点击此处可从《地质与资源》浏览原始摘要信息 |
| 点击此处可从《地质与资源》下载免费的PDF全文 |
|
