We present our approach to use a symbolic representation for the forecasting of power production from renewable energy sources.
We explain the basic principles of different symbolic regression methods, with a focus on sparse methods and evolutionary algorithms.
For comparison with physics-based statistical methods, we study hyperparameter search strategies and show how to combine both approaches in a hybrid model.
Data are taken from operating power plants. The results are shown in terms of normalized errors with corresponding discussion of sources of uncertainties.