Scholarlink

The excess emission of carbon dioxide (CO2) to the atmosphere has raised significant concerns about climate change. Consequently, industries with high CO2 emissions are under pressure to reduce their carbon footprint and explore solutions to minimise net CO2 output. One promising approach is carbon capture and storage (CCS), which involves storing liquefied CO2 underground. Assessing subsurface porosity is crucial for determining the feasibility of CCS projects as it helps to evaluate carbon storage capacity. In this study, we utilised three recently developed boosting algorithms—histogram-based boosting regression (HGBR), light gradient boosting machine regression (LGBR), and categorical boosting regression (CBR) to estimate subsurface porosity using well log data. We employed 5 well log data types: caliper log (CAL), gamma-ray log (GR), neutron porosity log (NPHI), photoelectric factor log (PE), and deep laterolog (LLD), as input features, while lab-corrected porosity served as the target variable. Model optimisation was conducted using grid search optimisation technique. Our results indicated that HGBR outperformed the other models, achieving an impressive coefficient of determination (R2) value of 0.9756. However, both LGBR and CBR also yielded high-performing models, with R2 values of 0.9598 and 0.9700, respectively. Among the input features, GR had the most significant influence, while PE had the least influence on the output. In conclusion, our findings suggest that all three boosting algorithms—HGBR, LGBR, and CBR—show promise for predicting porosity in sandstone layers using well logs, making them valuable tools for CCS assessment programs.

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

Application of Recently Developed Boosting Ensemble Machine Learning Algorithms in Carbon Capture and Storage Feasibility Assessment to Predict Subsurface Porosity

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Kushan Sandunil,
  • Ziad Bennour,
  • Saaveethya Sivakumar,
  • Hisham Ben Mahmud,
  • Ausama Giwelli

摘要

The excess emission of carbon dioxide (CO2) to the atmosphere has raised significant concerns about climate change. Consequently, industries with high CO2 emissions are under pressure to reduce their carbon footprint and explore solutions to minimise net CO2 output. One promising approach is carbon capture and storage (CCS), which involves storing liquefied CO2 underground. Assessing subsurface porosity is crucial for determining the feasibility of CCS projects as it helps to evaluate carbon storage capacity. In this study, we utilised three recently developed boosting algorithms—histogram-based boosting regression (HGBR), light gradient boosting machine regression (LGBR), and categorical boosting regression (CBR) to estimate subsurface porosity using well log data. We employed 5 well log data types: caliper log (CAL), gamma-ray log (GR), neutron porosity log (NPHI), photoelectric factor log (PE), and deep laterolog (LLD), as input features, while lab-corrected porosity served as the target variable. Model optimisation was conducted using grid search optimisation technique. Our results indicated that HGBR outperformed the other models, achieving an impressive coefficient of determination (R2) value of 0.9756. However, both LGBR and CBR also yielded high-performing models, with R2 values of 0.9598 and 0.9700, respectively. Among the input features, GR had the most significant influence, while PE had the least influence on the output. In conclusion, our findings suggest that all three boosting algorithms—HGBR, LGBR, and CBR—show promise for predicting porosity in sandstone layers using well logs, making them valuable tools for CCS assessment programs.