Short-term solar radiation forecast based on ensemble learning of multi-objective optimization

Accurate prediction of solar radiation is essential for the efficient use of photovoltaic energy. In this paper, a new ensemble learning model of multi-objective optimization is proposed to improve the performance of solar radiation prediction and the objectives are horizontal and directional accuracy. Firstly, the data of solar radiation is decomposed into a series of signal sets by Singular Spectrum Analysis (SSA). Then, the Least Squares Support Vector Machine (LSSVM) optimized by Non-dominated Sorting Genetic Algorithm (NSGAII) is used to predict each component signal. Moreover, clustering method is used to cluster the samples of each component signal. Finally, the NSGAII-LSSVM method is used to ensemble the sample results to get the final prediction value. In the paper the solar radiation data of Italy 2017 is used to verify the performance of the hybrid model. The results of the comparison with eight benchmark models show that the hybrid model has better performance and smaller error values than other benchmark models. In conclusion, the proposed ensemble learning model of multi-objective optimization is considerably effective and robust for the short-term solar radiation forecast.