Forecasting Solar Energy with Numerical Weather Prediction Data

Abstract

The thesis aims to predict total daily incoming solar energy at 98 Oklahoma Mesonet sites. These sites are considered as the solar farms. Prediction models are built using numerical weather prediction data from the NOAA/ESRL Global Ensemble Forecast System (GEFS) Reforecast Version 2. Data include 11 ensemble members with perturbed initial conditions and the forecast time steps 12, 15, 18, 21 and 24. There are as many as 15 model variables describing Numerical Weather Predictions (NWP). The idea is to explore various ways of utilizing the data and identifying suitable machine learning or statistical technique for prediction. The objective also includes experimenting with ensemble of predictors and finding the best way to aggregate predictions from the members of the ensemble. As many as seventeen different approaches are proposed and evaluated with evaluation measures such as Root Mean Square Error (RMSE), Normalized Root Mean Square Error (NRMSE), Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE).