Comparación de funciones kernel para la predicción de la oferta energética fotovoltaica

Recently, at the fields of climate change and energy demand have turned their attention to the study and discovery of patterns in renewable energies, such as the photovoltaic-type. Such patterns can be obtained by extrapolating radiation based on the electromagnetic spectrum bands captured by NASA’s Landsat and MODIS satellites, where artificial neural network (ANN) and support vector machine (SVM) algorithms have produced the best models. Nonetheless, the acquisition of training data from those sources is expensive, as well as it lacks the exploration of kernel functions for this application. Therefore, in this study, adjustments were made in the above aspects, mainly through: coupling of new kernels to ANN and SVM in the scikit-learn library, contributing to the reuse and robustness of these algorithms; and implementing an experimental framework to tune hyper-parameters, thus generating results comparable to those reported in the state of the art.