摘要
为了解决多源复杂测井曲线难以获得且生成困难这一难题,本文提出了一套高效的机器学习建模范式,即以长短期记忆神经网络为基础,通过数据完整性分析、交叉检验和收敛性分析进行建模。该建模范式主要有3个优点:可以提升训练数据的数量和质量、对模型性能进行评估并针对应用场景对模型的适用性进行预估。为了验证本文提出的建模范式的效果,以长宁威远地区页岩气井为例开展实验,结果显示该建模范式能够高效生成多源复杂测井曲线,且预测值与真实值具有相同的变化趋势,可以作为后续开发的参考。本文提出的建模范式易于使用且具有通用性,通过规范化建模流程有效降低机器学习建模的难度,有利于推广机器学习方法在测井曲线生成问题中的应用。
In order to solve the problem that multi-source complex log curves are difficult to be obtained and generated,this paper proposes an efficient machine learning modeling paradigm,that is,to establish a model on the base of long-short term memory,and through data integrity analysis,cross validation and convergence analysis.The modeling paradigm has three major advantages:it can improve the quantity and quality of training data,evaluate the performance of the model,and predict the applicability of the model for application scenarios.In order to verify the effect of the modeling paradigm proposed in this paper,shale gas wells in Changning-Weiyuan area are taken as examples to carry out experiments.The results show that the modeling paradigm can efficiently generate multi-source complex log curves,and the predicted values have the same change trend with real values,which can be used as a reference for subsequent development.The modeling paradigm proposed in this paper is easy to use and versatile.It can effectively reduce the difficulty of machine learning modeling by standardizing the modeling process,which is conducive to promoting the application of machine learning method in log curve generation.
作者
杨静
陈云天
蒋春碧
YANG Jing;CHEN Yuntian;JIANG Chunbi(Tangshan Vocational and Technical College,Tangshan,Hebei 063300,China;Peng Cheng Laboratory,Shenzhen,Guangdong 518000,China;Peking University,Beijing 100871,China)
出处
《中国海上油气》
CAS
CSCD
北大核心
2021年第1期76-84,共9页
China Offshore Oil and Gas
基金
“十三五”国家科技重大专项课题“页岩气藏地球物理响应与优质储层识别(编号:2017ZX05035-003)”部分研究成果。
