Optimización avanzada de la modelación de transporte de solutos en redes de alta complejidad
Cargando...
Autores
Silva Vargas, Luis Fernando
Director
Tipo de contenido
Document language:
Español
Fecha
Título de la revista
ISSN de la revista
Título del volumen
Documentos PDF
Resumen
En esta investigación se solucionó el problema inverso de flujo y transporte en un medio geológico de baja permeabilidad, usando como modelo conceptual las redes de fracturas discretas (DFN por sus siglas en inglés). Se realizó en un macizo fracturado real ubicado en el centro de España, donde se contaba con una adecuada caracterización geológica, datos de pruebas hidráulicas (de bombeo y de recuperación) y ensayos de trazadores conservativos (Deuterio y Uranina). Se utilizaron 77 redes de fracturas discretas percoladas. Se definió una metodología de calibración con el acople del modelo numérico TRANSIN IV y tres algoritmos de calibración automática: i) Algoritmo de máxima verosimilitud (Método de Levenberg Marquardt [Alcolea et al., 2000; Medina et al., 2004], ii) Shuffled complex evolution – University of Arizona (SEC-UA) [Duan et al., 1993], y. iii) Estimación generalizada de incertidumbre por verosimilitud GLUE (por sus siglas en inglés) [Beven y Binley, 1992]. En la solución del problema inverso para flujo, el 40% de las DFNs percoladas obtuvieron resultados satisfactorios y se realizó un análisis estadístico y de sensibilidad a los parámetros estimados, encontrando una enorme variabilidad de los parámetros en toda la red fracturas. En la solución del problema inverso para transporte, ninguna red percolada obtuvo resultados aceptables en el ajuste de las curvas de llegada para los dos trazadores, esto puede ser debido a que no se cumple el criterio de Péclét o la falta de conectividad de fracturas en la red. Entonces se utilizaron las redes sin percolar utilizadas por Donado [2009], en donde solo 5 redes obtuvieron resultados satisfactorios, mejorando el ajuste de las curvas sobre todo en las colas.
Abstract In this research, the inverse problem for flow and transport was solved in low permeable highly fractured massifs. These are studied using the Discrete Fracture Network (DFN) approach, this methodology can be used to analyze hydraulic (pumping and recovery) and tracer tests (Deuterium and Uranine) in a real fractured massif located in Central Spain. 77 different realizations of a percolated DFN were used in the calibration process. Calibration methodology is defined with the coupling numerical model TRANSIN IV and three automatic calibration algorithms: i) The Maximum Likelihood Method (The Levenberg-Marquat’s Method) [Alcolea et al., 2000; Medina et al., 2004], ii) Shuffled complex evolution – University of Arizona (SEC-UA) [Duan et al., 1993], y. iii) Generalized Likelihood Uncertainty Estimation GLUE [Beven y Binley, 1992]. In the inverse problem of flow, around 40% of the networks provide an excellent fitting in terms of heads. A statistical and sensitivity analysis was also done to the fitted parameters (Hydraulic conductivity and Storativity), finding a great variability of flow parameters across the fracture network. In the inverse problem for transport, no percolated network obtained acceptable results in the breakthrough curve for the two conservative tracers. this may be because the Péclét number is not satisfied in most of the calibrated meshes. Then the networks used without percolating used by Donado [2009], where only 5 networks provide an excellent fitting in terms of concentration in breakthrough curve, especially in the tails.
Abstract In this research, the inverse problem for flow and transport was solved in low permeable highly fractured massifs. These are studied using the Discrete Fracture Network (DFN) approach, this methodology can be used to analyze hydraulic (pumping and recovery) and tracer tests (Deuterium and Uranine) in a real fractured massif located in Central Spain. 77 different realizations of a percolated DFN were used in the calibration process. Calibration methodology is defined with the coupling numerical model TRANSIN IV and three automatic calibration algorithms: i) The Maximum Likelihood Method (The Levenberg-Marquat’s Method) [Alcolea et al., 2000; Medina et al., 2004], ii) Shuffled complex evolution – University of Arizona (SEC-UA) [Duan et al., 1993], y. iii) Generalized Likelihood Uncertainty Estimation GLUE [Beven y Binley, 1992]. In the inverse problem of flow, around 40% of the networks provide an excellent fitting in terms of heads. A statistical and sensitivity analysis was also done to the fitted parameters (Hydraulic conductivity and Storativity), finding a great variability of flow parameters across the fracture network. In the inverse problem for transport, no percolated network obtained acceptable results in the breakthrough curve for the two conservative tracers. this may be because the Péclét number is not satisfied in most of the calibrated meshes. Then the networks used without percolating used by Donado [2009], where only 5 networks provide an excellent fitting in terms of concentration in breakthrough curve, especially in the tails.