Predicción del recurso solar diario mediante técnicas de machine learning para la proyección de generación de energía eléctrica

Abstract

In this paper we design a model composed of Machine Learning techniques for the prediction of solar radiation in a daily time horizon. This model works under the supervised learning type conformed by Naive-Bayes and Neural Networks algorithms. The main objective of this work was to predict the daily solar resource by means of Machine Learning techniques for the projection of electric power generation in Colombia. The characteristics of the IDEAM and NASA databases were reviewed, the latter being the one selected to feed the input to the model due to its strengths over IDEAM for this study. The variables used to make the prediction are: Temperature, Relative Humidity, Wind Speed, Precipitation, Atmospheric Pressure, Clarity Index and Global Horizontal Solar Radiation. Four zones in the main regions of the Colombian territory, Soledad, Villavicencio, Buenaventura and Paipa, were used for training and testing the model. Statistical metrics such as RMSE, MSE, MAE, Coefficient of Determination, Covariance and Standard Deviation allowed the validation of the model, where the zone of Villavicencio is the one with the best performance with an RMSE of 0.02749 and Coefficient of Determination of 0.97138 and the zone with the lowest performance was Soledad with RMSE and Coefficient of Determination values of 0.04786 and 0.89258 respectively