A semi-empirical approach using gradient boosting and k-nearest neighbors regression for GEFCom2014 probabilistic solar power forecasting

The forecasting methods were gradient boosting for the deterministic forecasting of solar power and k-nearest neighbors regression for estimating prediction intervals in order to provide probabilistic forecasts. A cross-validation strategy, splitting the data into monthly folds, was employed for comparing the performances of alternative methods and in an attempt to avoid overfitting issues.