In this master's thesis, we investigated the possibility of using a simulation model developed in the Simulink environment to generate synthetic data for training machine learning models, namely XGBoost and Random Forest. The models were used to predict the minimum water level and the arrival time of the double flow in a chain of two small hydropower plants. Synthetic data obtained from the simulation model, as well as measurements from the actual system, were used for training and evaluating the models. The results show that synthetic data are suitable for predicting the minimum water level and enable a satisfactory prediction of its response. However, the results for predicting the arrival time of the double flow were not sufficiently reliable, as the simulation model does not yet describe this phenomenon accurately enough. The findings demonstrate the potential of using simulation data while also highlighting the need for further improvement of the simulation model.
|
