Modelo de predicción climática

The objective of this degree work was to develop a climate prediction model for weather forecasting in Canton Bolivar, it was necessary to use techniques such as the review of the state of the art, data collection and analysis. The information and results of other research were summarized, with the purpose of being up to date with the contributions made within the last five years about the object of study. Then, using a comparative analysis within the state of the art, it was determined that the most feasible model for its application within the work was the Recurrent Neural Network in conjunction with the structure of the long short- term memory or LSTM algorithm. Next, historical climate information pertaining to the area was collected, and for training and testing the model, records from the ESPAM - MFL meteorological station and the meteorological website Power Data Access Viewer (NASA) were used, and a data exploration and Pearson correlation was also performed to observe how related the variables of the datasets are, determining that they have independent behaviors among them. Finally, the model was built using the Python programming language in version 3.7.5, 150 epochs were defined for training, obtaining better results in the variables of humidity, maximum temperature and minimum temperature, with an accuracy between the real and predicted values of 89.27%, 92.00%, and 90.61% with the data from the meteorological station and with the second dataset they obtained 93.75%, 94.54%, and 96.96%.

• Autor principal: Kuffó Zambrano, Marjorie Stefany (author)
• Outros Autores: Pinargote Zambrano, Juan José (author)
• Formato: bachelorThesis
• Idioma: spa
• Publicado em: 2021
• Assuntos: Modelo de predicción; Pronóstico del clima; Series de tiempo climáticas; Aprendizaje automático
• Acesso em linha: http://repositorio.espam.edu.ec/handle/42000/1568