Tesis - Ingeniería en Electricidad

Permanent URI for this collection

https://repositorio.utc.edu.ec/handle/27000/5307

Browse

Browsing Tesis - Ingeniería en Electricidad by Subject "ÁRBOL DECISIÓN"

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Diseño de una herramienta de predicción mediante machine learning para la generación fotovoltaica en usuarios urbanos y rurales
    (Ecuador: Latacunga: Universidad Técnica de Cotopaxi: (UTC), 2024-08) Lojano Navas Cristofer Sergio; Castillo Fiallos, Jessica Nataly
    This research work responds to the electricity generation need because of the drought problems caused by lengthy droughts and climatic phenomena—El Niño, negatively affecting hydroelectric generation. Therefore, it is crucial to explore new sources of renewable energy. Photovoltaic systems emerge as an alternative, given that the sun is an accessible source and its use for electricity generation represents an accessible alternative. For this research project, solar radiation data is collected from 2017 to 2013 in Tabacundo, which is appropriately processed and purified to ensure its quality. In addition, the database is indexed and then divided into sets, using the years 2017-2022 and part of 2023 for training (80%) and the rest of 2023 for validation (20%). The DecisionTreeRegressor library allows the algorithm to be trained and predicted in Python software. The decision tree model is reliable for short-term predictions with a precision of 0.976, sensitivity of 0.998, accuracy of 0.974 and R^2 of 0.935. However, its long-term performance is poor, obtaining MAE results of 8.694 in 1 month and 139.6 for 6 months. It is recommended to consider other models such as LSTM and more variables to improve precision. For photovoltaic sizing, 15 panels are required for a consumption of 160 kWh in grid-connected systems and 17 in isolated systems, resulting in higher costs due to the need for more components and a continuous supply, being less advisable.