Medellín - Minas - Maestría en Ingeniería - Analítica · 2025
Prediction of crude oil-water interfacial tension with surfactants and nanomaterials using machine learning
Accurate prediction of interfacial tension (IFT) is a critical factor for the design and optimization of chemical enhanced oil recovery (cEOR) processes. This study focuses on the application of four predictive models (RF, ET, GBRT, XGBoost) for IFT in systems with surfactants and nanomaterials. For this purpose, an experimental and literature-based dataset with 551 datapoints was used, characterized by an imbalanced distribution composed of 75% IFT measurements below 20.55 mN·m⁻¹ (systems with additives) and 25% from control experiments with higher values. This data structure was intentionally preserved to ensure the model's phenomenological representativeness. Model performance was evaluated using performance metrics (R², RMSE, MAE), residual plots, and learning curves. Analysis of the learning curves revealed that the model's performance stops improving after approximately 200 training samples, demonstrating that incorporating similar additional data is not beneficial. The results confirm that the random forest model is the most robust tool for predicting IFT with an R² of 85% and underscore that a representative data composition is more crucial than a strict statistical balance, offering valuable guidance for optimizing future data collection efforts. (Texto tomado de la fuente)
Texto completo 57 páginas con texto de 62
Leer la tesis completa Ficha en el repositorio
Contenido
- Introductionp. 17
- Motivationp. 17
- Problem statementp. 19
- Objectivesp. 20
- 1.3.1 General Objectivep. 20
- 1.3.2 Specific Objectivesp. 20
- Thesis Scopep. 21
- Document Structurep. 21
- Theoretical frameworkp. 22
- Surfactantsp. 43
- 2.1.1 Bagging Learningp. 24
- 2.1.2 Boosting Learningp. 25
- Phenomena Related to the Interfacial Tension of Nanofluids containing Surfactantsp. 26
- Systematic literature reviewp. 30
- Nanofluidsp. 38
- Surfactant-Modified Nanofluidsp. 32
- Nanomaterials with Machine Learningp. 38
- Variable Selection in Interfacial Tension Prediction using Machine Learningp. 39
- Interfacial Tension Data Compilation: Experimental and Literature Sourcesp. 41
- Nanomaterialsp. 20
- Nanomaterialsp. 20
- containing Surfactantsp. 43
- Resultsp. 45
- Interfacial Tension Measurements of Nanofluids with Surfactantsp. 45
- Analysis and Description of the Dataset's Structurep. 47
- Performance of the Implemented Machine Learning Modelsp. 49
- Conclusions and Future Workp. 56
- Referencesp. 58
- Figure 1-1. Colombia's Total Energy Supply for 2023. Adapted from [2]p. 17
- Figure 1-2. Evolution of Energy Supply in Colombia since 2000. Adapted from [2]p. 18
- Figure 2-1. Classification of Machine Learning Approaches. Adapted from [28]p. 23
- Figure 2-2. Ensemble Learning: Bagging & Boosting techniques. Adapted from [30]p. 24
- literature reviewp. 32
- reduced interfacial tensionp. 40
- Figure 5-1. IFT vs Concentration CQDp. 46
- Figure 5-2. IFT vs Concentration CQD in the presence of Surfactantp. 47
- Figure 5-3. Dataset distributionp. 48
- Figure 5-4. Spearman correlation matrix for the continuous variablesp. 49
- Figure 5-5. Performance metricsp. 50
- Random Forest and (B) Extra Treesp. 51
- Gradient Boost Regressor Tree and (B) Extreme Gradient Boostp. 52
- Random Forest, and (B) Gradient Boosting and XGBoostp. 54
- Table 3-1 Research equations used for the systematic review of the literaturep. 31
- Table 3-2. Predictive Models and Performance Metrics in IFT Prediction Related Studiesp. 36
- Table 4-1. Hyperparameters used to train the modelsp. 44
- Table 5-1. Descriptive Statistics of Interfacial Tension in the Datasetp. 48
- Table 5-2. Tuned Model Hyperparametersp. 55