A New Closed-Loop Solar Power Forecasting Method With Sample Selection

In this article, a new short-term solar power forecasting method is proposed which has a closed-loop structure composed of point-estimating and range-classifying parts. If the forecasts generated by these parts for solar power are inconsistent, the feedback loop sends appropriate signals to them to correct their predictions. The feedback loop iterates until consistent forecasts are generated for the solar power by the point-estimating and range-classifying parts. This enables the proposed closed-loop forecasting method to enhance its solar power prediction accuracy and reliability. Furthermore, a novel sample selection approach, different from feature selection methods, is devised to mine the historical data for finding the most informative training samples for training the proposed forecasting engine. The effectiveness of the proposed solar power forecasting method is illustrated by testing it on some real-world solar farms and comparing its results with the results of several state-of-the-art solar power prediction methods.