基于混合集成学习模型的测井曲线生成方法
Well Log Generation Based on Hybrid Ensemble Learning Model
王宵宇 1廖广志 1肖立志 1黄文松 2孔祥文 2赵子斌2
作者信息
- 1. 油气资源与工程全国重点实验室,中国石油大学(北京),北京 102249
- 2. 中国石油勘探开发研究院,北京 100083
- 折叠
摘要
测井曲线在储集层评价和油气资源评估中具有十分重要的作用,但是实际应用中经常出现部分测井曲线缺失的情况,而重新测井的成本高昂且实现困难.为了在不增加经济成本的基础上补充缺失的测井曲线,提出了一种基于混合集成学习模型的测井曲线生成方法,以高效智能的方式补全缺失的测井曲线.混合集成学习模型结合了随机森林模型和极限梯度提升模型的结构优势,深度挖掘测井数据之间的非线性映射关系,实现了对测井曲线的精准生成.将混合集成学习模型应用于实际测井数据,并将其生成结果与全连接神经网络模型和多元线性回归模型的生成结果进行对比分析,实验结果表明混合集成学习模型合成的人工测井曲线精度更高,说明了混合集成学习模型适用于生成测井曲线,为人工测井曲线合成提供了一种新的思路.
Abstract
Well logs play a crucial role in reservoir evaluation and oil and gas resource assessment.However,the shortage issue of some well logs always exists in practical applications.The well log data are remeasured through a drilling procedure involving high cost and difficult implementation.To supplement the missing well logs without increasing economic costs,this study develops an approach for generating well logs based on the hybrid ensemble learning model.This approach efficiently and intelligently completes the missing well logging curves.The hybrid ensemble learning model combines the structural advantages of random forest model and extreme gradient boosting model,enabling deep exploration of the nonlinear mapping relationships among well log data and achieving accurate predictive results of log wells.The proposed hybrid ensemble learning model is applied to real well log data,and the generated results are compared with those of fully connected neural network model and multivariate linear regression model.Experimental results show that the well logs synthesized by the hybrid ensemble learning model exhibit higher accuracy,indicating the suitability of the hybrid ensemble learning model for well log generation.This study provides a new perspective for synthesizing artificial well logs.
关键词
测井曲线/曲线生成/多元线性回归/全连接神经网络/混合集成学习Key words
well log/log generation/multivariate linear regression/fully connected neural network/hybrid ensemble learning引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据