摘要
针对岩心标定的隔夹层数据少且隔夹层与砂岩样本分布不均衡等问题,运用深度自编码器以及半监督学习方法,通过计算异常得分,并将异常得分情况赋予分类置信度,根据分类置信度得到隔夹层的分类结果,并对模型进行更新。研究结果表明,采用更新算法的深度自编码器模型在隔夹层识别中效果明显,综合分类准确率均达到了85.00%,且相较于其他分类算法,最优模型AE7&UP的F1_(score)最高,为84.15%,说明模型的识别效果好且均衡。研究成果对重构地下流体认知体系具有重要意义。
In order to solve the problems such as few interlayer data calibrated by core and unbalanced distribution of interlayers and sandstone samples,the deep self-encoder and semi-supervised learning method were used to calculate the abnormal scores and give the classification confidence to the abnormal scores.According to the classification confidence,the classification results of the interlayers were obtained,and the model was updated.The research results show that the deep self-encoder model using the update algorithm is effective in the interlayer identification,and the comprehensive classification accuracy reaches 85.00%.In addition,compared with other classification algorithms,the optimal model of AE7&UP has the highest F1_(score)(84.15%),indicating that the identification effect of the model is good and balanced.The research results are of great significance to reconstruct the cognition system of underground fluids.
作者
陈雁
焦世祥
程超
黄成
蒋裕强
Chen Yan;Jiao Shixiang;Cheng Chao;Huang Cheng;Jiang Yuqiang(Southwest Petroleum University,Chengdu,Sichuan 610500 China)
出处
《特种油气藏》
CAS
CSCD
北大核心
2021年第1期86-91,共6页
Special Oil & Gas Reservoirs
基金
国家自然科学基金“基于特征提取与分层建模的社交网络信息传播预测研究”(61503312)
四川省科技厅项目“海相页岩气建产核心区智能评价系统研究(省重)”(19YYJC)
南充市市校科技战略合作项目“智能舆情分析技术研究”(18SXHZ0010)
西南石油大学2018年高等教育教学改革研究一般项目“新工科背景下的嵌入式课程‘AI+’教学新模式的探索与实践”(X2018JGYB041)。
关键词
隔夹层识别
自编码器
深度学习
半监督
interlayer identification
self-encoder
deep learning
semi-supervision
