Title : Development of a high-resolution Total Organic Carbon (TOC) prediction model in shale gas reservoirs using continuous well log data
Abstract:
Objective of the Project:
The project aims to develop a predictive model for estimating the Total Organic Carbon (TOC) content in shale gas reservoirs. Traditionally, TOC has been assessed through laboratory methods, such as core analysis, which can be limited in providing continuous high-resolution information. The project seeks to leverage continuous well log data to overcome these limitations and contribute to a more comprehensive understanding of source rock properties.
Introduction:
The growing interest in unconventional hydrocarbon production from shale source rocks underscores the importance of key petrophysical properties like Total Organic Carbon (TOC). TOC serves as a critical indicator of the quality of source rocks, influencing hydrocarbon quantification and resource quality assessment. Traditional laboratory methods for TOC quantification, such as Rock-Eval 6 pyrolysis, while reliable, can be time-consuming and costly, especially in thick reservoirs with lateral variations. To address these challenges, utilizing well log data has become essential, with various approaches attempting to estimate TOC. However, existing methods may fall short in accuracy due to the involvement of multiple parameters and the heterogeneity of shale formations.
Methodology:
The project employs a comprehensive methodology for predictive modeling:
1. Data Cleaning and Scaling: Ensuring the quality and consistency of well log data.
2. Data Analysis: Thorough analysis of well log data to identify patterns and correlations.
3. Train Test Split: Division of the dataset into training and testing sets for model development and evaluation.
4. Developing the Model: Implementation of a predictive model, with a focus on Random Forest regression.
5. Obtaining Assessment Metrics: Evaluation of model performance using Root Mean Square Error (RMSE).
6. Parameters Analysis In-depth analysis of parameters influencing the predictive model.
7. Comparison of Results: Comparative assessment of different regression models.
Observation:
Diverse regression models were applied to the predictive model, revealing distinct mean squared error values. Notably, the Random Forest Regression and Extra Tree Regression models demonstrated superior performance with remarkably low errors of 0.0644 and 0.0609, respectively. This comparative analysis underscores the efficacy of ensemble-based techniques for precise Total Organic Carbon (TOC) prediction in shale gas reservoirs, surpassing the accuracy of traditional methods such as Linear Regression and Support Vector Regression.
Conclusion and Future Plans:
The project concludes that the Random Forest regressor is the best-fit model for predicting TOC values, achieving a minimal Root Mean Square Error of 0.0644. This finding supports the successful prediction of TOC using well log data. In the future, the project aims to extend its application by predicting CO2 sequestration potential based on the TOC values obtained.
This research offers a valuable contribution to reservoir characterization, enabling more accurate and cost-effective assessments of TOC in shale gas reservoirs using continuous well log data.
Novelty/Additive Information:
The project introduces a novel feature selection technique, incorporating advanced machine learning algorithms to enhance predictive accuracy. Additionally, it pioneers the integration of well log data for predicting Total Organic Carbon, marking a significant advancement in efficient and precise reservoir characterization compared to conventional laboratory methods.
Audience Takeaway Notes:
- Practical Implementation: Apply the presented methodology for developing predictive models in shale gas reservoir projects, improving efficiency over traditional lab methods.
- Efficient Data Utilization: Enhance reservoir characterization by effectively using continuous well log data, offering time and resource savings.
- Model Selection Guidance: Select appropriate regression models, with emphasis on ensemble-based techniques like Random Forest, for accurate TOC prediction.
- Research Expansion for Faculty: Serve as a foundation for faculty interested in expanding research in predictive modeling and reservoir characterization.
- Simplified Decision-Making for Designers: Provide a practical solution to streamline TOC assessment, supporting informed decision-making and design efficiency.
