6DE537F1-26BF-4E5F-BDB0-2D07AA89FFA3.pptx

Great content

1. APPLICATION OF MACHINE LEARNING ALGORITHMS FOR IMPROVED PREDICTION OF PERMEABILITY By: Group 13 Supervisor: Dr. Jonathan Atuquaye Quaye Petroleum Engineering Faculty of Civil & Geo Engineering

2. Outline Introduction Aims and objectives Methodology Results and Discussion Conclusion Recommendation

3. Introduction Permeability is the measure of a reservoir rocks ability to allow fluids to flow through when subjected to a pressure gradient. Conventional practical methods have proven to be time and cost-consuming and can take up to two years to ensure convergence of predicted values for permeability(Lindsay et al., 2023) . In this study, we are embarking on the capability of three machine learning algorithms for improved prediction of permeability.

4. Aims and Objectives Aim: The main aim of this study is to study the integration of machine-learning algorithms in the determination of permeability and select a machine-learning algorithm that enhances the prediction of permeability in hydrocarbon reservoirs. Objectives: To make a comparative study between the various supervised machine learning algorithms. To select the most effective algorithm based on parameters such as accuracy, the execution time to predict permeability and the error margin.

5. Methodology Exploratory Data Analysis Model selection & optimization Performance analysis Data collection and definition Result interpretation End

6. Data Collection and Definition For this research, 76 core plugs were collected from the field and tested in the lab. Special and routine core analysis tests were performed on the plugs to evaluate rock properties such as pore volume, porosity and permeability.

7. Exploratory Data Analysis (EDA) EDA involves summarizing and understanding the main characteristics of a dataset using visual methods. For this research, components of EDA utilized were data visualization, descriptive statistics, data cleaning and feature engineering.

8. Descriptive Statistics Descriptive statistics measures the dataset’s distribution, central tendency and variability.

9. Data Visualization

10. Data Cleaning Outliers were handled by using robust algorithms. SVR handles outliers with epsilon ( ε ) and a regularization parameter (C) to minimize error margins. DT handles them by creating splits that isolate the few outliers to minimize their effect on the accuracy. The various features of the dataset were also subjected to standardization.

11. Feature Engineering This was used to select the features with the highest influence on the permeability. Sample length, pore volume, porosity and volumetric density were identified. Feature engineering was carried out using univariate selection which utilizes f-regression statistics.

12. Model Selection & Optimization Linear regression was first used on the data to serve as a simple baseline against which more complex models could be compared. Decision tree and support vector regressions were then utilized and optimized using gridsearch to find the best possible combinations of hyperparameters that result in the best performance of the models. Algorithm Hyperparameter Decision Tree Max_depth: None, Min_samples_leaf: 2, Min_samples_split: 10 Support Vector Regression 'C': 100, 'epsilon': 0.001, 'gamma': 'scale'

13. Performance Analysis Performance of the models were evaluated based on four statistical metrics . MSE RMSE MAE Linear Regression 0.1937 0.4401 0.2995 0.9054 Decision Tree 0.0307 0.1751 0.1219 0.9850 Support Vector Regression 0.0207 0.1439 0.1147 0.9899 MSE RMSE MAE Linear Regression 0.1937 0.4401 0.2995 0.9054 Decision Tree 0.0307 0.1751 0.1219 0.9850 Support Vector Regression 0.0207 0.1439 0.1147 0.9899

14. Results and Discussion

15. Conclusion SVR outperforms other algorithms : Superior accuracy and efficiency compared to Decision Trees and Linear Regression. Effective use of machine learning algorithms can reduce the time needed to evaluate permeability on the field. SVR's success points to further research opportunities with hybrid models and advanced techniques.

16. Recommendation Data from other fields should be integrated to enhance the model’s predictive ability across various basins with different rock properties. Integrate SVR into reservoir analysis for improved accuracy and faster predictions, enhancing reliability and efficiency.

17. THANK YOU