Research

The estimation of energy production losses due to residual or environmental flows was carried out around 20 years ago. That assessment was based on simplified assumptions, especially regarding the hydrology and small hydropower. The improved availability of data, as well as findings from measures already implemented, should now help to provide a more accurate estimate of generation losses. The analyses will not only update the figures from the previous, but also determine the effects of gradual increases in residual flow until the full implementation of the WFD.

One of the important tasks of Austrian Power Grid AG (APG) is the medium-term forecast of energy production from hydropower plants in order to be able to coordinate and plan the availability and utilization of the Austrian electricity grid accordingly. The objective of the planned project is to improve the medium-term forecast (24, 48 and 72 hours) for the "medium" small hydropower plants. An initial focus will be placed on the Salzburg model region, where the data situation for the development and implementation of AI forecasting methods is very good. This initial feasibility study will be divided into several phases: In consultation with the client, initially two AI methods (XGBoost, and Long-Short-Term-Memory (LSTM) models) will be tested and optimized with the VTW's as the sole input variables for the locations of the CHP plants. In further steps, static catchment area characteristics (topography, soil, geology, climate, vegetation) are integrated into the methods as additional information. External drivers such as precipitation and weather forecasts from ZAMG/GeoSphere or model-based estimates of the so-called snow-water equivalent of the existing snow cover are then taken into account.

The aim of the project is to develop an improved methodology for the semi-automated age structure assessment of the BQE fish in flowing waters in accordance with Guideline A1 on the basis of fish surveys in flowing waters and machine learning processes. Data sets from more than 4,000 surveys from the "national water status monitoring" and other data from projects that can be used to assess the ecological status of fish are currently available throughout Austria. A core element of the assessment, which is otherwise automated, is the "assessment of the length-frequency diagrams" (= assessment of the age structure), which must be carried out on the basis of an expert assessment and therefore also represents a weak point. Experienced colleagues have tried to automate this step, but have so far only been successful in some areas. As part of the project, methods from the machine learning (ML) field, such as "XGBoost" or "Random Forests", are now to be used and adapted to carry out an assessment of the length-frequency diagrams based on fish caught and measured in a section.