Scholarlink

<p>For shale gas development, it is essential to clarify the main controlling factors of EUR and realize its accurate prediction. Based on data from 123 wells in the Weiyuan block, this study combines Pearson correlation analysis, RF-RFE algorithm, and Boruta algorithm to screen out the main controlling factors of EUR. The GBDT, RF, SVR, MLP, and ERT algorithms optimized by the Optuna framework are then used to build EUR prediction models, and the optimal model is further interpreted using SHAP. The results show that: (1) Five main controlling factors are selected: drilled length of type I reservoir, thickness of the L1<sub>1</sub><sup>1</sup> sub-member, fracturing fluid volume, vertical depth, and 360-day flowback rate. (2) Optuna-GBDT outperforms other models on both the training and test sets, with an <i>R</i><sup>2</sup> of 0.8214 on the test set, demonstrating high prediction accuracy. (3) The global feature importance ranking based on the mean absolute SHAP values of each feature is as follows: thickness of the L1<sub>1</sub><sup>1</sup> sub-member &gt; drilled length of type I reservoir &gt; 360-day flowback rate &gt; fracturing fluid volume &gt; vertical depth. The first four features contribute positively to the model output to different degrees, while vertical depth has a more complex effect on the model in this study, not showing a simple positive or negative trend. The Optuna-GBDT model, combined with SHAP-based interpretation, provides a practical and efficient idea for EUR prediction in shale gas wells.</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

Interpretable machine learning for predicting EUR of shale gas wells in the Weiyuan block

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月11日

查看全文
  • Sijie He,
  • Feifei Fang,
  • Jin Chang,
  • Xizhe Li,
  • Ruilan Luo,
  • Nijun Qi

摘要

For shale gas development, it is essential to clarify the main controlling factors of EUR and realize its accurate prediction. Based on data from 123 wells in the Weiyuan block, this study combines Pearson correlation analysis, RF-RFE algorithm, and Boruta algorithm to screen out the main controlling factors of EUR. The GBDT, RF, SVR, MLP, and ERT algorithms optimized by the Optuna framework are then used to build EUR prediction models, and the optimal model is further interpreted using SHAP. The results show that: (1) Five main controlling factors are selected: drilled length of type I reservoir, thickness of the L111 sub-member, fracturing fluid volume, vertical depth, and 360-day flowback rate. (2) Optuna-GBDT outperforms other models on both the training and test sets, with an R2 of 0.8214 on the test set, demonstrating high prediction accuracy. (3) The global feature importance ranking based on the mean absolute SHAP values of each feature is as follows: thickness of the L111 sub-member > drilled length of type I reservoir > 360-day flowback rate > fracturing fluid volume > vertical depth. The first four features contribute positively to the model output to different degrees, while vertical depth has a more complex effect on the model in this study, not showing a simple positive or negative trend. The Optuna-GBDT model, combined with SHAP-based interpretation, provides a practical and efficient idea for EUR prediction in shale gas wells.