Data driven models for renewable energy and load forecasting

The progressively advanced diffusion of energy from renewable sources leads to an increasingly articulated networks and their more complex management. Among of the main issues to be addressed there are the aleatory of renewable energy sources (RES), how to predict the power they feed into the grid and how to match all the different types of current energy production. In addition, both electrical and thermal load curves have been shaped to new lifestyles and consequently bring with them greater difficulties in being able to forecast this loads. Thus from these themes this thesis is born with the aim of being able to use more innovative methods in making predictions. These methods adopted are based on machine learning and artificial neural networks applying data driven models. Therefore researches were done to find forecasting methods using artificial neural networks. Next, two renewable energy sources and a load curve were selected in order to fine-tune these forecasting techniques. The RES selected are a photovoltaic plant with data collected online, and a solar thermal collector system with data provided by SOLID Solar Energy Systems GmbH, Graz, Austria. Eventually the load curve forecast concerns a district heating system located in Alba, Italy. For each part, results were presented regarding the accuracy of the models built.