Modeling solar radiation and wind speed time series for renewable energy applications

The problem of predicting weather variables, such as solar radiation and wind speed, is of great interest for integrating renewable energies plants, into the electric grid. Indeed, since renewable energy sources are intermittent in nature, predicting future values is important to allow the grid to dispatching generators, in order to satisfy the demand. There are essentially two ways to address the issue of weather variables prediction. One is by using Numerical Weather Forecasting (NWF) models, which are reliable, but also quite complex and requires real time information, usually available from Meteorological Agencies only. Furthermore, very powerful computers are required to solve the differential equations involved. The other kinds of methods are represented by the so-called statistical modeling approaches, which are based on the use of past data recorded at the site of interest. These latter kinds of methods, compared to the former ones, require less computational efforts, but are appropriate only for short time horizons. This PhD Thesis was devoted to study short-term prediction models for solar radiation and wind speed time series and assessing their performance in the range [1, 24] hours. It was also studied the predictability of the daily average values, which for obvious reasons, is much more difficult than that of predicting the hourly averages. To mitigate, as far as possible, the difficulties, the prediction was reformulated in terms of a classification problem. In such a way, instead of predicting 1-day ahead the average value, the target was to predict the class. In this framework, of course, the prediction is as far difficult as large is the number of considered classes. Thus accuracy of 1-day ahead prediction models of the wind speed class was studied, for various frameworks