Métodos estadísticos y modelos de Machine Learning para la predicción de la radiación solar en Iquitos, 2023

• Autores: Tirado Tello, César, Dulanto Torres, Christian Alberto
• Formato: tesis de grado
• Fecha de Publicación: 2023
• Institución: Universidad Nacional De La Amazonía Peruana
• Repositorio: UNAPIquitos-Institucional
• Lenguaje: español
• OAI Identifier: oai:repositorio.unapiquitos.edu.pe:20.500.12737/9979
• Enlace del recurso: https://hdl.handle.net/20.500.12737/9979
• Nivel de acceso: acceso abierto
• Materia: Redes neuronales (informática); Estadística matemática; Análisis de series temporales; Predicción; Radiación solar; https://purl.org/pe-repo/ocde/ford#2.02.04; https://purl.org/pe-repo/ocde/ford#1.01.03

This investigative project deployed a meticulous comparison between two methods for predicting solar radiation in Iquitos, utilizing the statistical ARIMA model and a machine learning model grounded in neural networks, with a mixed non-experimental design approach and data provided by SENAMHI, encompassing variables such as solar radiation, relative humidity, and atmospheric pressure. Despite the neural network model displaying formidable performance, accounting for 98.85% of the variability in the training dataset, a decline in its efficiency was observed in the validation and testing phases, hinting at potential overfitting. Conversely, the ARIMA model, while showcasing merely acceptable performance with an MSE of 45.651 and an R² of 0.8500, demonstrated steady robustness. The comparison elucidated that, although neural networks possess superiority in predictive capacity, explaining 95.51% of the variability and showcasing a lower mean squared error, it is imperative to weigh other critical elements such as model interpretation and computational resources when finalizing decisions. Ultimately, while machine learning methods, particularly neural networks, proved to be especially insightful in predicting solar radiation, the exploration and optimization of different modeling approaches emerge as an essential horizon for future research in the field.