Numerical strategies for characterizing fractured rock from heat tracer experiments

Characterization of fractured rocks is a key challenge for optimizing heat harvesting in geothermal systems. The use of heat as a tracer, facilitated by the development of such advanced techniques as active line source (ALS) borehole heating and the distributed temperature sensing (DTS), shows the great potential for characterizing fractured rocks. However, there is so far a limited number of theoretical and numerical studies on how these tests could be used for estimating both fracture-network and rock-matrix properties.

We use deep neural networks to describe heat tracer test data and demonstrate how the cumulative density function (CDF) or probability density function (PDF) of the heat tracer test data can be deployed in the inversion mode, i.e., to infer the fracture parameters with. Our approach utilizes the methods of distributions, developed previously to estimate the CDF of solute concentration described by a reactive transport model with uncertain parameters and inputs. The method is applied to analyze several synthetic heat tracer test datasets obtained from a particle-based forward model of transport processes in heterogeneous fractured rocks. The study considers alternative representations of fracture networks with a large range of variation of the fracture network properties, as well as several experimental conditions (e.g., ambient/forced thermal and hydraulic conditions, pulse/continuous changes in temperature). This allows us to characterize the system by combining the information from several thermal tests.