Deep learning and geochemical modelling as tools for solute geothermometry

Geothermometry is constituted as one of the most important geochemical tools for reservoir exploration and development. Solute geothermometers are used to estimate the temperature in the subsurface. Therefore, the chemical composition of a discharging geothermal fluid is used to infer the temperature of the reservoir. Conventional solute geothermometers are using element ration or single mineral phases for reservoir temperature estimation. These geothermometers have large uncertainties applying them on the same fluid sample. Therefore, a new method called multicomponent geothermometry was introduced. Assuming the reservoir rock and the geothermal fluid are in equilibrium, the mineral assemblage of the reservoir rock is used to determine the temperature within the reservoir. Thus, the temperature-dependent saturation indices of the mineral set are calculated. Utilising multiple mineral phases for temperature estimation leads to a statistically more robust and precise result than the conventional method. Nevertheless, secondary processes such as boiling, degassing, and dilution are disturbing the equilibrium reaction within the fluid.
... mehr
Hence, we developed a solute multicomponent geothermometer with an additional optimisation process. Geochemical modelling improves the methodology. Several interdependent, vulnerable parameters especially pH value, aluminium concentration, as well as boiling, and dilution are considered. These parameters are elaborated in the geochemical modelling process to optimise these values to fit their in-situ reservoir conditions again. Therefore, the uncertainty within the temperature estimation is furthermore reduced. However, this process is time-consuming, and geochemical as well as mineralogical knowledge would be beneficial for applying the method.
On the other hand, artificial intelligence is a powerful tool to solve complex issues, even considering unknowns. Therefore, we developed a new adequate solute geothermometer based on an artificial neuronal network. Selected geochemical fluid parameters are used as input parameters. Further on, the net was trained and validated with a high-quality dataset. The best performing net was tested and compared to conventional and multicomponent geothermometers as well as in-situ temperature measurements of geothermal wells. Concluding in a neuronal solute geothermometer as precise as our optimised multicomponent geothermometer but much easier and faster in its applicability.