-
Felipe P. Marinho
Programa de Pós-Graduação em Engenharia de Teleinformática, Universidade Federal do Ceará, CE
-
Ajalmar R. R. Neto
Programa de Pós-Graduação em Ciências da Computação, Instituto Federal do Ceará, CE
-
Paulo A. C. Rocha
Departamento de Engenharia Mecânica, Universidade Federal do Ceará, CE
Keywords:
Envelope Estimation, Multiple Linear Regression, Machine Learning, Solar energy, Solar Irradiance Forecast
Abstract
In this work, we employ the recent envelope estimation method to fit a multiple linear regression model in order to make predictions of solar irradiance at horizons of 10, 20 and 30 min ahead. The advantage of using this method lies in the fact that there is a reduction in the variance of the estimated coefficients. Such a reduction is obtained by enveloping the material part, while excluding the variation of the immaterial part, which also contributes to a reduction in dimensionality, since there is a decrease in the estimation variance without an increase in the number of observations. The model was used in a dataset formed by LDR luminosity signals and statistical attributes extracted from images of the sky. The integration of the LDRs and the camera was done using a Raspberry Pi 3. Its performance was compared to that of linear (LASSO and Ridge) and non-linear (Multi-Layer Perceptron) models, being evaluated by the metrics Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Root Mean Square Error Relative (rRMSE).
