Pronóstico de radiación solar mediante aprendizaje automático utilizando software de código abierto

The present research work focuses on predicting solar radiation. The importance of making this prediction lies in its various applications such as: renewable energy generation, photosynthesis for plant growth, for thermal systems for water heating, etc. The methodology used involves the design, implementation and comparison of several prediction models, these models are based on machine learning techniques and are implemented using open source software (Python), among the techniques used are simple linear regression, closed recurrent units (GRU) and neural networks. To evaluate the effectiveness of these models, metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE) and the coefficient of determination (R2) are used to measure the accuracy of the predictions compared to the real data. For the analysis of the results obtained from each prediction model, it is carried out in three scenarios: one day (2023/02/06), one week (2023/02/13 to 2023/02/19) and one month (March 2023), in which, the one that stands out best in the three analysis scenarios is the simple linear regression model due to its consistent superior performance in several metrics. With a low Mean Absolute Error (MAE = 19.379), high Coefficient of Determination (R2 = 0.9937) close to 1, and the lowest Mean Squared Error (MSE = 525.963) in all instances, this technique proves to offer more accurate predictions and a robust fit.

• Päätekijä: Muguicha Hinojoza, Cristhian Ismael (author)
• Aineistotyyppi: bachelorThesis
• Kieli: spa
• Julkaistu: 2024
• Aiheet: APRENDIZAJE AUTOMÁTICO; PRONOSTICO; MÉTRICAS; MODELOS; ELECTRICIDAD
• Linkit: http://repositorio.utc.edu.ec/handle/27000/11983