The intention of the present thesis is to create a model for short-term forecasting (i.e. up to 72 hours ahead), using auto-learning techniques, of Photovoltaic power plants. The main objective consists in applying innovative concepts based on Extreme Learning Machines (ELM) with variable coefficients, allowing the modelization of non-linear relations between the power plant production and Numerical Weather Predictions (NWP).

 

The experimentation with several different activation functions is also an integrant part of this thesis in order to understand its effects on the results and to discover which activation function provides better results.