Evaluación De Un Modelo Predictivo De La Viscosidad En Emulsiones De Agua En Crudo (W/O) Basado En Aprendizaje Automatizado (Machine Learning): Efecto De La Distribución Del Tamaño De Gota Y El Contenido De Agua
Cargando...
Fecha
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
Las emulsiones de agua en crudo (W/O) pueden impactar sensiblemente la producción de un campo petrolero al incrementar costos y reducir el flujo de la producción. El efecto de la distribución del tamaño de gota (DSD) en el comportamiento reológico de una emulsión no se ha establecido claramente. Para abordar esta necesidad, el presente estudio evalúa la utilidad de un modelo predictivo basado en Machine Learning para estimar la viscosidad de emulsiones W/O, utilizando datos experimentales de DSD y contenido de agua (% vol.). Se desarrolló un modelo de red neuronal densa con hiperparámetros optimizados mediante un algoritmo genético estándar. El desempeño del modelo se evaluó mediante validación cruzada k-fold, con métricas MSE, MAE, RMSE y R². Aunque el modelo alcanzó una precisión aceptable en algunos conjuntos de muestras y en relación con modelos del ámbito de la producción de hidrocarburos, la capacidad predictiva fue limitada (MSE: 0.27, R²: 0.52) debido a una baja correlación entre DSD y viscosidad. A pesar de que el modelo no generaliza adecuadamente, los hallazgos son de gran utilidad en etapas exploratorias y diagnósticas. Se recomienda ampliar el conjunto de muestras e incorporar variables composicionales y de comportamiento de interfase.
Water-in-oil (W/O) emulsions can significantly impact oilfield production by increasing costs and reducing production flow. The effect of droplet size distribution (DSD) on the rheological behavior of an emulsion has not been clearly established. To address this need, the present study evaluates the usefulness of a Machine Learning-based predictive model to estimate the viscosity of W/O emulsions using experimental DSD and water content (% vol.) data. A dense neural network model with optimized hyperparameters was developed using a standard genetic algorithm. Model performance was evaluated by k-fold cross-validation, with MSE, MAE, RMSE and R² metrics. Although the model achieved acceptable accuracy on some sample sets and relative to models in the hydrocarbon production domain, the predictive capability was limited (MSE: 0.27, R²: 0.52) due to a low correlation between DSD and viscosity. Although the model does not generalize adequately, the findings are useful in exploratory and diagnostic stages. It is recommended to expand the sample set and incorporate compositional and interphase behavior variables.
Water-in-oil (W/O) emulsions can significantly impact oilfield production by increasing costs and reducing production flow. The effect of droplet size distribution (DSD) on the rheological behavior of an emulsion has not been clearly established. To address this need, the present study evaluates the usefulness of a Machine Learning-based predictive model to estimate the viscosity of W/O emulsions using experimental DSD and water content (% vol.) data. A dense neural network model with optimized hyperparameters was developed using a standard genetic algorithm. Model performance was evaluated by k-fold cross-validation, with MSE, MAE, RMSE and R² metrics. Although the model achieved acceptable accuracy on some sample sets and relative to models in the hydrocarbon production domain, the predictive capability was limited (MSE: 0.27, R²: 0.52) due to a low correlation between DSD and viscosity. Although the model does not generalize adequately, the findings are useful in exploratory and diagnostic stages. It is recommended to expand the sample set and incorporate compositional and interphase behavior variables.