Predicción de nubes a corto plazo para una plataforma solar a partir de datos radiométricos
Loading...
Official URL
Full text at PDC
Publication date
2018
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
Internet de las cosas es un paradigma que ha revolucionado la conexión entre las personas y los objetos generando en tiempo real una gran cantidad de datos. Debido a esta revolución, diversos campos están viviendo un gran aumento en su utilización, y entre ellos se encuentra el campo de las energías renovables. En concreto, la energía solar está teniendo una velocidad de desarrollo muy acentuada, necesitando nuevas formas de actuar y de gestionar las instalaciones. En este trabajo se aborda el problema de la predicción de radiación global sobre superficie horizontal con alta resolución espacial y temporal (5 minutos) a partir de los datos registrados durante un año en la red radiométrica de alta resolución ubicada en la Plataforma Solar de Almería (PSACIEMAT 1 ). En particular se muestra un método capaz de predecir el valor de radiación en los siguientes minutos a partir de los valores de los minutos anteriores. El método emplea el tipo de red neuronal recurrente conocido como LSTM, capaz de aprender patrones complejos y predecir el próximo elemento de una serie temporal. Los resultados muestran una mejora apreciable en la precisión del método con respecto a la predicción basada en el último valor conocido.
The Internet of Things is a paradigm that has revolutionized the connection between people and objects, generating a large amount of data in real time. Due to this revolution, diverse fields are experiencing a large increase in their use, and among them is the field of renewable energies. In particular, solar energy has a very high development speed, new ways of acting and managing facilities are needed. This work deals with the problem of the prediction of global radiation on a horizontal surface with high spatial and temporal resolution (5 minutes) from the data recorded during a year in the high resolution radiometric network locate in the Solar Platform of Almería (PSA-CIEMAT 2 ). In particular, a method capable of predicting the radiation value in the following minutes from the values of the previous minutes is shown. The method employs the type of recurrent neural red known as LSTM, capable of learning complex patterns and predicting the next element of a time series. The results show an appreciable improvement in the accuracy of the method with respect to the prediction based on the last known value.
The Internet of Things is a paradigm that has revolutionized the connection between people and objects, generating a large amount of data in real time. Due to this revolution, diverse fields are experiencing a large increase in their use, and among them is the field of renewable energies. In particular, solar energy has a very high development speed, new ways of acting and managing facilities are needed. This work deals with the problem of the prediction of global radiation on a horizontal surface with high spatial and temporal resolution (5 minutes) from the data recorded during a year in the high resolution radiometric network locate in the Solar Platform of Almería (PSA-CIEMAT 2 ). In particular, a method capable of predicting the radiation value in the following minutes from the values of the previous minutes is shown. The method employs the type of recurrent neural red known as LSTM, capable of learning complex patterns and predicting the next element of a time series. The results show an appreciable improvement in the accuracy of the method with respect to the prediction based on the last known value.
Description
Facultad de Informática, Departamento de Sistemas Informáticos y Computación, curso 2017-2018.