Scholarlink

<p>Porosity estimation at high granite temperatures is essential for numerous purposes, including natural and enhanced geothermal energy production. However, these methods of determining porosity have some disadvantages, such as being labor-intensive, requiring expensive instrumentation, and taking a significant amount of time, particularly during elevated temperature treatments. This study is vital for advancing geothermal energy applications by addressing the limitations of traditional porosity estimation methods. The study introduces innovative predictive machine learning (ML) models and offers insights into practical applications for improving geothermal reservoir management and sustainability. the datasets divided into six subsets, each with a specific grain size (fine, medium, or coarse) and cooling method furnace cooling (FC) and water cooling (WC). The cooling rate has a significant impact on rock thermal stress. In this study, six ML models—Random Forest (RF), K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM)—were employed to predict models for estimating porosity at high temperatures. The multiple evaluation metrics were used to assess the suitability of these algorithms. R-squared values were calculated to assess the models’ goodness-of-fit. In parallel, several error metrics such as root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square relative error (RMSRE), mean absolute relative error (MARE), and percent bias (PBIAS) were evaluated to quantify the accuracy of the predictions. The performance of the ML methods exhibited considerable variability among the different datasets. Among the evaluated models, CatBoost and KNN consistently achieved higher R-squared values and lower error metrics, demonstrating their effectiveness in accurately discerning underlying patterns and reliably predicting porosity. Conversely, RF, XGBoost, and SVM yielded reasonably accurate predictions, albeit with slightly increased variability in their performance metrics across different grain size and cooling condition subsets. LightGBM demonstrates comparatively diminished prediction accuracy, since it fails to capture the complex fluctuations in rock properties over thermal heating and cooling cycles. These findings highlight the superior predictive efficacy of CatBoost and KNN, confirming their reliability as robust tools for modeling porosity in complex datasets with varied geological and environmental factors.</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

Granite porosity prediction under varied thermal conditions using machine learning models

  • 期刊论文

  • 发表日期：

    2025年1月29日

查看全文
  • Rishabh Dwivedi,
  • Balbir Prasad,
  • PK Gautam,
  • Peeyush Garg,
  • Siddhartha Agarwal,
  • KH Singh,
  • TN Singh

摘要

Porosity estimation at high granite temperatures is essential for numerous purposes, including natural and enhanced geothermal energy production. However, these methods of determining porosity have some disadvantages, such as being labor-intensive, requiring expensive instrumentation, and taking a significant amount of time, particularly during elevated temperature treatments. This study is vital for advancing geothermal energy applications by addressing the limitations of traditional porosity estimation methods. The study introduces innovative predictive machine learning (ML) models and offers insights into practical applications for improving geothermal reservoir management and sustainability. the datasets divided into six subsets, each with a specific grain size (fine, medium, or coarse) and cooling method furnace cooling (FC) and water cooling (WC). The cooling rate has a significant impact on rock thermal stress. In this study, six ML models—Random Forest (RF), K-Nearest Neighbors (KNN), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Categorical Boosting (CatBoost), and Light Gradient Boosting Machine (LightGBM)—were employed to predict models for estimating porosity at high temperatures. The multiple evaluation metrics were used to assess the suitability of these algorithms. R-squared values were calculated to assess the models’ goodness-of-fit. In parallel, several error metrics such as root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), root mean square relative error (RMSRE), mean absolute relative error (MARE), and percent bias (PBIAS) were evaluated to quantify the accuracy of the predictions. The performance of the ML methods exhibited considerable variability among the different datasets. Among the evaluated models, CatBoost and KNN consistently achieved higher R-squared values and lower error metrics, demonstrating their effectiveness in accurately discerning underlying patterns and reliably predicting porosity. Conversely, RF, XGBoost, and SVM yielded reasonably accurate predictions, albeit with slightly increased variability in their performance metrics across different grain size and cooling condition subsets. LightGBM demonstrates comparatively diminished prediction accuracy, since it fails to capture the complex fluctuations in rock properties over thermal heating and cooling cycles. These findings highlight the superior predictive efficacy of CatBoost and KNN, confirming their reliability as robust tools for modeling porosity in complex datasets with varied geological and environmental factors.