Evaluating machine learning in short-term forecasting time series of solar power

Abstract

Alternative sources of energy are becoming more and more frequent, aiming to reduce environmental pollution, besides being ideal to overcome the energy crisis, therefore, in this context, solar power stands out for being abundant. Due to the high level of uncertainty of the factors that directly interfere in the generation of solar power, such as temperature and solar radiation, making solar energy predictions with high precision is a challenge. Thus, the objective of this article is to develop a forecast model through time series that makes it possible to predict the production of solar power, for 1, 3 and 6 steps ahead, emphasizing the potential of the neural network, using a database of one photovoltaic plant located in Uruguay. For the development of the proposal, pre-processing techniques and forecasting methods Support Vector Regression (SVR), Bayesian Regularized Neural Network (BRNN) and Generalized Linear Model (GLM) were combined. Finally, these combinations were compared using performance measures. It was noted that the combination of principal components analysis (Principal Components Analysis - PCA) and the Multilayer Perceptron Neural Network with Bayesian Regularization obtained the best results, using the three performance measures.