物探化探计算技术2025，Vol.47Issue(3)：410-419,10.DOI:10.12474/wthtjs.20240318-0001
基于CL-Trans模型的测井储层参数预测方法
Prediction method for reservoir parameters based on CL-Trans network
胡睿 1李勇 1刘应天 1冯文1
作者信息
- 1. 成都理工大学地球物理学院,成都 610059
- 折叠
摘要
Abstract
Effective reservoir parameters are important for characterization and evaluation.Traditional methods based on core measurements or petrophysical modeling are either expensive or inefficient,and existing methods based on recurrent neural networks cannot sufficiently capture the global dependencies in the sequences.In this paper,a reliable,low-cost method for reservoir parameter prediction that is sensitive to the global features of logging data is established from actual logging data.The technique is a hybrid deep learning model based on CNN,LSTM,and Transformer encoder,called CL-Trans network.The CL-Trans network firstly uses one-dimensional convolutional layers to mine potential primary features from logging data,then utilizes multiple LSTM layers to establish nonlinear relationships between logging data and reservoir parameters,and finally applies it to a Transformer encoder with a self-attention mechanism to further extract global features from logging data.We used this network to a region of logging data,and predicted porosity and permeability,showing more accurate and stable reservoir parameter prediction results when compared with Random Forest,LSTM,and CNN-LSTM network.关键词
机器学习/Transformer编码器/地球物理测井/储层参数预测Key words
machine learning/transformer encoder/Geophysical logging/prediction of reservoir parameters分类
地球科学引用本文复制引用
胡睿,李勇,刘应天,冯文..基于CL-Trans模型的测井储层参数预测方法[J].物探化探计算技术,2025,47(3):410-419,10.
