Thesis APLICACIÓN DE REDES RECURRENTES PARA LA PREDICCIÓN DE POTENCIA EN PLANTAS EÓLICAS CHILENAS
Date
2018
Authors
VALENZUELA CARRASCO, CAMILO ANDRES
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
El uso de Energías Renovables no Convencionales (ERNC) en Chile va aumentando año a año,pero la incorporación de estas energías a un Sistema Eléctrico Nacional pensado para la generaciónde energía tradicional no es simple. La energía eólica es una las ERNC que ha tomado mayorrelevancia el último tiempo, no obstante que el comportamiento estocástico de la velocidad delviento hace difícil predecir la potencia eólica generada en el futuro, por lo que es necesario tenermodelos de predicción eficientes a distintos horizontes de tiempo.En este trabajo se realizará una comparación entre los modelos estadísticos tradicionales lineales,con los modelos del estado del arte de las Redes Neuronales Recurrentes (RNN). Mostrando que lasRNN obtienen mejores resultados a distintos horizontes de predicción gracias a que son modelosaltamente no lineales y no realizan supuestos de la serie al momento de modelar.
The use of Non-Conventional Renewable Energy (NCRE) in Chile is increasing every year, butthe incorporation of these energies into a National Electric System designed for traditional energygeneration methods is not simple. Wind power is the NCRE that has become more relevant in recenttimes, although the stochastic behavior of wind speed makes it difficult to predict the wind powergenerated in the future, so it is necessary to have efficient prediction models at different horizons.In this paper a comparison between the traditional linear statistical models, with the state of theart models of the Recurrent Neural Networks (RNN) will be made. Showing that the RNN obtainbetter results at different prediction horizons thanks to the fact that they are highly non-linear modelsand do not make assumptions of the series at the moment of modeling.
The use of Non-Conventional Renewable Energy (NCRE) in Chile is increasing every year, butthe incorporation of these energies into a National Electric System designed for traditional energygeneration methods is not simple. Wind power is the NCRE that has become more relevant in recenttimes, although the stochastic behavior of wind speed makes it difficult to predict the wind powergenerated in the future, so it is necessary to have efficient prediction models at different horizons.In this paper a comparison between the traditional linear statistical models, with the state of theart models of the Recurrent Neural Networks (RNN) will be made. Showing that the RNN obtainbetter results at different prediction horizons thanks to the fact that they are highly non-linear modelsand do not make assumptions of the series at the moment of modeling.
Description
Catalogado desde la version PDF de la tesis.
Keywords
ARIMA , ENERGIA EOLICA , ENERGIAS RENOVABLES , MACHINE LEARNING , REDES NEURONALES RECURRENTES