Abstract
Population growth and continuous demand for energy production have prompted electricity producers to focus increasingly on the use of local and clean resources. Renewable energy sources such as the sun and wind have shown an increasing interest. A problem with these power sources are their intermittent and random nature that drives grid operators to restrict their integration into the energy mix. This makes it imperative to combine different production systems to ensure grid stability and safety. French Guiana’s electricity production mainly comes from local renewable energies (64%), compared to 36% from imported fossil fuels. To improve the integration of intermittent renewable energies, solar energy in this study, it is necessary to focus on resource forecasting. Knowing in advance the available power allows optimal management of the coupling between conventional and intermittent production systems.The contribution of this thesis focuses on forecasting the Global Horizontal Irradiation (GHI) at different time horizons combining satellite-derived data instead of ground measurements and statistical methods. Satellite images offer the advantage of providing irradiance products over wide areas and with satisfactory accuracy. We are interested in forecasting the GHI as the generated PV power directly depends on the incoming GHI intensity. In this thesis, we have developed and studied ten solar radiation prediction models. Their methods are the Persistence, Scaled Persistence, AR, ARMA, Gaussian Process, Support Vector Machine, simple regression trees, bagged forests, WRF with Kalman filter, and an aggregation method. Satellites-derived data from the meteorological geostationary satellite GOES-13 are used as input of each model. These models were first developed to predict the GHI using satellite-derived data as input, and then to predict the GHI using ground measurements to be able to quantify discrepancies. The forecast horizons tested are from 1 to 6H per hour time step, to provide useful horizons for network managers. Data from six measurement sites located in French Guiana are used for GHI prediction. Five years of data from the six stations were used for the learning phase and one year for the validation phase.