Scholarlink

<p>Based on continuous production data from 112 wells in a deep shale gas block, comprising approximately 160,000 production days, this study proposes a hybrid deep learning framework integrating the Rabbit Optimization Algorithm (ROA), Transformer, and Mamba architectures for daily shale gas production forecasting. Average casing pressure, daily water production rate, and flowback ratio are employed as input features, while ROA is used to globally optimize model hyperparameters. Model performance and robustness are evaluated using five-fold cross-validation. compared with the standalone Mamba model, the proposed ROA-Transformer–Mamba framework reduces RMSE from 0.0418 to 0.0328 (approximately 21.5%) and MAE from 0.02438 to 0.0174 (approximately 28.6%), while increasing the coefficient of determination to 0.938., demonstrating superior prediction accuracy and generalization capability. To the best of the authors’ knowledge, this is the first study to integrate ROA, Transformer, and Mamba architectures for shale gas production forecasting, providing an effective data-driven solution for capturing complex production dynamics under multi-factor coupling conditions.</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

Forecasting deep shale gas production using a ROA-optimized Transformer–Mamba hybrid network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月3日

查看全文
  • Weikang He,
  • Xizhe Li,
  • Yujin Wan,
  • Nan Wang,
  • Honming Zhan,
  • Xiangyang Pei,
  • Longyi Wang,
  • Wenxuan Yu,
  • Yuhang Zhou

摘要

Based on continuous production data from 112 wells in a deep shale gas block, comprising approximately 160,000 production days, this study proposes a hybrid deep learning framework integrating the Rabbit Optimization Algorithm (ROA), Transformer, and Mamba architectures for daily shale gas production forecasting. Average casing pressure, daily water production rate, and flowback ratio are employed as input features, while ROA is used to globally optimize model hyperparameters. Model performance and robustness are evaluated using five-fold cross-validation. compared with the standalone Mamba model, the proposed ROA-Transformer–Mamba framework reduces RMSE from 0.0418 to 0.0328 (approximately 21.5%) and MAE from 0.02438 to 0.0174 (approximately 28.6%), while increasing the coefficient of determination to 0.938., demonstrating superior prediction accuracy and generalization capability. To the best of the authors’ knowledge, this is the first study to integrate ROA, Transformer, and Mamba architectures for shale gas production forecasting, providing an effective data-driven solution for capturing complex production dynamics under multi-factor coupling conditions.