Hybrid approaches based on Singular Spectrum Analysis and [formula omitted]- Nearest Neighbors for clearness index forecasting

Abstract

With the increase in solar energy penetration in electric power systems, it is necessary to predict future solar radiation. With this, the solar power generation capacity can be predicted, allowing accurate scheduling for other sources. The solar radiation or the clearness index (CI) prediction was already presented using artificial intelligence, hybrid algorithms, among other approaches. In this paper, two hybrid approaches for clearness index forecasting are proposed. The first one combines batch singular spectral analysis and k-nearest neighbors (k-NN), and is called SSA-B + k-NN, whereas the second associates recursive singular spectral analysis and k-NN, and is named SSA-R + k-NN. The first one performs the SSA-B algorithm for the time series decomposition using a more usual approach, in which the eigenvalues and eigenvectors are calculated in batch. The second one performs the SSA-R algorithm for the time series decomposition, in which the eigenvalues and eigenvectors are updated recursively . The computational costs and mean square errors (MSE) were compared. The results were also compared to the ones obtained with other methods based on k-NN published in the literature. The results illustrate that the hybrid approaches proposed in this paper obtained a lower MSE compared to the results previously presented.