Forecasting Solar Radiation Strength Using Machine Learning Ensemble

To enhance the forecasting of solar radiation strength on horizontals, an ensemble learning approach is proposed. Two types of machine learning models are arranged to predict solar radiation, namely the recurrent neural networks, and the support vector regressors. A multi-layer perceptron model acting as an ensemble learning technique is deployed to combine the forecasts of ensemble models through a stacking technique. The combiner provides an automatic weighted averaging of the forecasters' outputs. The suggested approach helps improving the accuracy of a one-day-ahead solar energy forecasting. The performance of the combining technique is evaluated over an entire year using meteorological data. The experiments showed the superiority of the learning-based combinatory model compared to those of single models as well as to other combining techniques.