Scholarlink

<p>Porosity, permeability, and water saturation in oil reservoirs are key factors in evaluating whether potential exploration areas are suitable for development, which directly affect drilling speed, oil and gas production, and development costs. Due to the complex nonlinear relationship between reservoir parameters, accurate prediction is crucial. Firstly, this paper presents a deep learning hybrid model, DBOCM-BiTCN-GRU-CA, designed to predict evaluation parameters from complex formation depth data. The model extracts and combines features from BiTCN-GRU (a neural network tailored for time series tasks) and enhances its ability to capture both global and short-term dependencies by incorporating the Cross-Attention (an advanced attention mechanism). At the same time, the Dung Beetle Optimization algorithm with Chebyshev mapping (DBOCM) is used to optimize network hyperparameters to enhance the model's stability and accuracy. Secondly, 9811 sample data points, ranging in depth from 2248.8 to 3229.8 m, are selected for prediction. The input variables include natural gamma (GR), deep laterolog resistivity (RD), acoustic travel time (DT), micro lateral resistivity (RML), spontaneous potential (SP), and shallow laterolog resistivity (RS). Finally, experiments comparing the DBOCM-BiTCN-GRU-CA model with XGBoost and PSO-BiLSTM demonstrate its superior predictive accuracy. The results show that the mean absolute error (MAE) and root mean square error (RMSE) for effective porosity and water saturation predictions are both below 2%, while those for permeability prediction are below 6%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Research on prediction method of reservoir key parameters using deep network architecture based on cross feature fusion with optimization mechanism

  • 期刊论文

  • 发表日期：

    2025年4月8日

查看全文
  • Fengcai Huo,
  • XingLin Zhao,
  • Hongli Dong,
  • Weijian Ren,
  • Qinjiang Sun

摘要

Porosity, permeability, and water saturation in oil reservoirs are key factors in evaluating whether potential exploration areas are suitable for development, which directly affect drilling speed, oil and gas production, and development costs. Due to the complex nonlinear relationship between reservoir parameters, accurate prediction is crucial. Firstly, this paper presents a deep learning hybrid model, DBOCM-BiTCN-GRU-CA, designed to predict evaluation parameters from complex formation depth data. The model extracts and combines features from BiTCN-GRU (a neural network tailored for time series tasks) and enhances its ability to capture both global and short-term dependencies by incorporating the Cross-Attention (an advanced attention mechanism). At the same time, the Dung Beetle Optimization algorithm with Chebyshev mapping (DBOCM) is used to optimize network hyperparameters to enhance the model's stability and accuracy. Secondly, 9811 sample data points, ranging in depth from 2248.8 to 3229.8 m, are selected for prediction. The input variables include natural gamma (GR), deep laterolog resistivity (RD), acoustic travel time (DT), micro lateral resistivity (RML), spontaneous potential (SP), and shallow laterolog resistivity (RS). Finally, experiments comparing the DBOCM-BiTCN-GRU-CA model with XGBoost and PSO-BiLSTM demonstrate its superior predictive accuracy. The results show that the mean absolute error (MAE) and root mean square error (RMSE) for effective porosity and water saturation predictions are both below 2%, while those for permeability prediction are below 6%.