The project focused on improving the efficiency of the geothermal power plant in Traunreut by investigating adiabatic cooling, parameter optimisation of the overheating temperature, and the use of machine‑learning models to evaluate process changes. The main technical outcome of the adiabatic cooling study was a simulation result from VEE that indicated a potential increase of about 500 MWh, corresponding to 1.64 % of the plant’s annual electricity production. A feasibility assessment of the required water withdrawal was positive, but a comprehensive risk analysis concluded that, under the existing operating conditions, the retrofit of adiabatic cooling would not be economically justified.
The overheating‑temperature optimisation was carried out in close cooperation with the plant manufacturer Turboden. The process parameters were lowered, and the change was tested in normal operation. Although the expected additional production was achieved, it came at the cost of a significant rise in the plant’s self‑consumption. Consequently the optimisation was abandoned and the original settings were restored. These results underline the difficulty of improving performance by small parameter changes when the plant’s operating envelope is already tightly constrained.
To overcome the limitations of thermodynamic simulation, a machine‑learning (ML) approach was introduced. Fraunhofer IGCV built a data‑driven model of the Traunreut plant using operational data. The model was then used to predict the impact of various process modifications. The ML model could quantify the benefit of a partial cleaning of the heat‑exchanger surface with dry ice compared to the standard water cleaning, but it could not reliably predict the effect of changing the blade‑angle settings. The same modelling effort was applied to a proposed replacement of the refrigerant R134a with a lower global‑warming‑potential alternative. The economic analysis showed that the costs of replacing and disposing of the existing refrigerant would outweigh any operational savings, leading to the rejection of this option.
Parallel to the modelling work, the project upgraded the measurement infrastructure. Compact PT100 thermometers were installed on all 24 condensate‑line outlets of the condensers to record the temperature of the working fluid after condensation. Four additional thermometers were mounted on the fan hoods to capture the exhaust‑air temperature. A capacitive humidity sensor was added to provide data for the adiabatic‑cooling design. All sensor data are recorded by the plant’s process‑control system and archived in a central database. Attempts to measure the flow of the working fluid in the condensers using ultrasound were made, but the welded piping configuration limited the accuracy of the measurements, so this approach was not pursued further.
The collaboration involved several partners. GKT, the operator of the Traunreut geothermal plant, provided the plant data and operational support. Turboden supplied the equipment and technical expertise for the parameter optimisation. Fraunhofer IGCV contributed the ML modelling and data‑analysis capabilities. Vulcan Energy Engineering (VEE) supplied the simulation tools and performed the adiabatic‑cooling feasibility study, while Vulcan Energy Ressourcen (VER) handled the project’s resource management. The project was carried out over a period of several years, funded through a German research programme aimed at enhancing the efficiency of geothermal power generation. The results demonstrate that while simulation and ML can identify potential gains, practical implementation must consider economic viability and the limits of measurement accuracy in existing plant infrastructure.