石油地球物理勘探2025，Vol.60Issue(3)：555-563,9.DOI:10.13810/j.cnki.issn.1000‑7210.20240201
嵌入注意力机制的时空网络设计及孔隙度可靠性预测
Design of spatio-temporal network embedded with attention mechanism and prediction of porosity reliability
摘要
Abstract
Porosity is an important indicator for evaluating reservoirs and calculating reserves.However,the traditional coring method is costly to obtain porosity,and the porosity prediction method based on regression analysis and a statistical model often has significant errors.Therefore,a reservoir porosity prediction model that combines convolutional neural network(CNN),bidirectional long short‑term memory network(BiLSTM),and attention mechanism is constructed,and its performance is verified using actual well logging data.Firstly,the complex nonlinear spatio‑temporal relationships of logging data are captured with CNN and BiLSTM.Then,the convolutional self‑attention mechanism is embedded,which generates queries and keys by causal convolution and allows better integration of local information into the attention mechanism.Compared with traditional self‑attention mechanisms,this approach avoids the influence of abnormal data on the prediction results.Finally,the Monte Carlo dropout approach is used to quantify the uncertainty of the model,providing confidence intervals for reservoir porosity prediction and further assessing prediction credibility.The compari‑son experiments among multiple models show that the proposed method has high accuracy in predicting reser‑voir porosity.The experiments on two wells with different characteristics show that the method has strong generalization ability.关键词
储层孔隙度预测/卷积神经网络/双向长短期记忆网络/注意力机制/不确定性量化Key words
reservoir porosity prediction/convolutional neural network/bidirectional long short‑term memory network/attention mechanism/uncertainty quantification分类
地质学引用本文复制引用
李艳辉,陶悦..嵌入注意力机制的时空网络设计及孔隙度可靠性预测[J].石油地球物理勘探,2025,60(3):555-563,9.基金项目
本项研究受河北省自然科学基金项目"通信资源受限情形下网络化非线性系统控制方法研究及其在油田注水系统中应用"(F2023107002)资助. （F2023107002）
