Improvement of short-term runoff forecast for the VERBUND hydro-power plant along Inn and Danube

The short-term forecast (1-4h) for the Inn and Danube power plants is an economically important planning instrument for VERBUNG AG. However, the current forecasts via the model system COSERO are not satisfactory in phases of rising and falling discharge regimes. In the project, we will try to significantly improve the short-term forecasts with the help of machine learning procedures using all available input data. Methods that will be adapted and further developed are stepwise linear regression, random forest, XGBoost and feed foreward neural network. The existing COSERO model runs and the classical "Adaptive gain transfer function" concept serve as Benschmark models. In a first phase, the potential of the method will be analyzed for 2 selected power plants.