Enfoque multicriterio para la selección óptima de variables explicativas para modelos de pronóstico de la energía eléctrica de plantas solares fotovoltaicas

Abstract

When a forecast problem is approached through regression models, it is expected to have the optimal number of explanatory variables and, if not, to be able to apply some technique to reduce the problem’s dimensionality. Currently, there is a variety of methods to select the features or explanatory variables, which in turn fall into different categories, making it complex to select only the ideal method for a specific application. Therefore, this research aims to present a multicriteria methodology for the optimal selection of the explanatory variables of a regression model, using the feature selection methods as the decision criteria and the explanatory variables as the alternatives. The methodology is illustrated through the data set of a photovoltaic solar plant from the National Institute of Standards and Technology (NIST) of the United States, taking the AC electricity generated by the plant as the objective variable and the temperature of the solar panels, the ambient temperature, and the wind speed as explanatory variables to solar irradiance. Methods of the "filter" type, the "wrapper" type, and the "embedded" type are considered. Using the TOPSIS multicriteria technique, it was possible to select the best variable to represent solar irradiance with a weighting of 1.00, the temperature of the solar panels of 0.182, the ambient temperature of 0.204, and the wind speed of 0.129.