A benchmark study for dynamic multilevel multiscale (ADM) simulation of heat production from low-enthalpy fractured geothermal reservoirs

Abstract

Accurate and efficient predictions on the behavior of fluid flow and heat transport are required in the development of low-enthalpy geothermal reservoirs in fractured formations. Key challenges include the demand for high-resolution computational grids, the non-linear behavior of the system due to strong mass-heat coupling and the presence of fractures with large heterogeneity contrasts. In this work, a comparison is made between natural and molar variable formulation used to describe the coupled fluid-heat transport under non-isothermal conditions in low-enthalpy fractured porous media. The solutions and performance of the newly implemented molar formulation are compared to those of the existing natural variable formulation in the DARSim2 reservoir simulation framework. A fully implicit scheme (FIM) is applied to solve the coupled discrete system including mass and energy balance equations. Application of the Algebraic Dynamic Multilevel (ADM) method with projection-based Embedded Discrete Fracture Model (pEDFM) provides a scalable and efficient simulation framework for field-scale fractured reservoirs. The ADM method maps the fine-scale system onto a dynamically defined multilevel grid resolution system (Cusini et al., 2016, HosseiniMehr et al., 2020) based on the solution gradient and a series of restriction and prolongation operators, which ensure accurate capturing of fine-scale heterogeneities. Fractures are defined explicitly as either (highly) conductive passageways or flow barriers using the pEDFM formulation (Tene et al., 2017). Simulation results using the molar formulation are compared with an analytical solution as verification of the implementation. Results of various (un-)fractured test cases with homogeneous and heterogeneous permeability fields show that there is no clear difference between the solutions and performance of the different primary variable formulations, and the performance itself is largely dependent on the level of complexity embedded in the numerical model independent of the simulation strategy applied.