石油地球物理勘探2025，Vol.60Issue(5)：1099-1110,12.DOI:10.13810/j.cnki.issn.1000-7210.20240319
基于卷积残差BiLSTM网络的层理缝定量表征方法
Quantitative characterization method of bedding fractures based on convolution residual BiLSTM network
摘要
Abstract
Bedding fractures are common in tight sandstone reservoirs and shale reservoirs.As the reservoir space and seepage channel of oil and gas,they have a significant impact on oil enrichment and production effi-ciency.The traditional prediction methods of bedding fractures is limited by the quality of seismic and logging data,as well as the number of actual drilling wells,with some limitations in accuracy and efficiency.In recent years,deep learning technology has been widely used in fracture identification and prediction,but with the in-crease in model complexity grows,the problems of gradient anomaly and performance degradation are beco-ming increasingly obvious,and the commonly used models fail to fully adapt to sequence seismic and logging data.Therefore,this paper proposes a new method for bedding fracture prediction,which is based on the convo-lutional residual bidirectional long short-term memory neural network(BiLSTM).Firstly,pseudo-wells are uniformly deployed in the study area to solve the problem that the number of actual drilling wells is insufficient and it is difficult to fully cover the study area.Combined with the core observation data,the well-side seismic at-tributes of a variety of real drilling wells and pseudo-wells with statistical information on bedding fractures are extracted to establish training samples and actual prediction data sets.Secondly,through the sample expansion and preprocessing related technical means are adopted to solve the problem of sample quality problem.Finally,the convolutional neural network is used to extract sample features,and the convolution residual connection is established to transmit data to the BiLSTM network with gating mechanism for information selection and forget-ting.This effectively alleviates the problems of gradient anomaly and performance degradation in the deep net-work,and significantly improves the prediction accuracy of the model,with the coefficient of determination reaching up to 91.3%.The prediction results of bedding fractures in the M region of Subei Basin show that the proposed method can relatively efficiently and accurately predict the development condition of bedding frac-tures,and the prediction results are consistent with geological understanding.This method provides effective support and practical guidance for on-site oil and gas exploration industry.关键词
层理缝/卷积残差/BiLSTM/地震属性/伪井/苏北盆地Key words
bedding fractures/convolution residual/BiLSTM/seismic attributes/pseudo-well/Subei Basin分类
天文与地球科学引用本文复制引用
圣学礼,胡慧婷,付晓飞,王海学,钱诗友,孙雅雄..基于卷积残差BiLSTM网络的层理缝定量表征方法[J].石油地球物理勘探,2025,60(5):1099-1110,12.基金项目
本项目受黑龙江省自然科学基金优秀青年基金项目"断层分段生长量化表征及对页岩油'输导'和'保存'作用研究"(YQ2022D007)、国家自然科学基金项目"断裂对页岩油富集控制作用研究——以松辽盆地南部青一段页岩油为例"(U20A2093)联合资助. （YQ2022D007）
