Publication: Estimación de intervalos de predicción mediante computación evolutiva
Loading...
Identifiers
Publication date
2015-09-23
Defense date
2015-10-09
Authors
Advisors
Tutors
Journal Title
Journal ISSN
Volume Title
Publisher
Impact
Export
Abstract
Con la creciente necesidad de energía en el mundo moderno, cada vez es más importante la utilización de una energía limpia y segura.La energía en la que nos centraremos en este do-cumento es la solar. Debido a las limitaciones de almacenamiento de este tipo de energía, es necesario el desarrollo de técnicas para poder predecir la electricidad generada. Se explicarán los métodos seguidos para la creación de modelos de predicción de intervalos basados en técnicas de inteligencia arti˝cial, como son las redes de neuronas. Esta red contará 300 neuronas de entrada con los atributos atmosféricos sobre radiación solar y densidad de las nubes, y con dos salidas para representar ambos extremos del intervalo. Además, se hará uso de técnicas de optimización para los pesos de la red, para lo cual utilizaremos algoritmos de computación evolutiva como los genéticos o los de evolución diferencial. Para ello, también se detallara la metodología usada para obtener los cromosomas y función de ˝tness necesarias. Por último, estos modelos se compararán con un método de construcción de intervalos, la regresión cuantil.
As the modern world's need for energy keeps growing, using a clean source of energy is more important every day. In this project we will focus on solar energy. Due to the limitations of this kind of energy in regard of the way it is stored, it is necessary to develop technics to predict the generated electricity. It will be explained the methods for the creation of models of prediction for intervals based on arti cial intelligence technics, such as neural networks. This network will have 300 inputs neurons that will specify the atmospheric attributes like solar radiation or cloud density, and two output neurons that will represent the extremes of the interval. Also, we will use optimization technics for the weights of the network, for which we will use evolutionary computation algorithms, such as genetics algorithms and di erential evolution algorithms. For this, it will be detailed the methodology used for the construction of the chromosomes and the tness function. At least, these models will be compared with an interval predictor method, the quantile regression.
As the modern world's need for energy keeps growing, using a clean source of energy is more important every day. In this project we will focus on solar energy. Due to the limitations of this kind of energy in regard of the way it is stored, it is necessary to develop technics to predict the generated electricity. It will be explained the methods for the creation of models of prediction for intervals based on arti cial intelligence technics, such as neural networks. This network will have 300 inputs neurons that will specify the atmospheric attributes like solar radiation or cloud density, and two output neurons that will represent the extremes of the interval. Also, we will use optimization technics for the weights of the network, for which we will use evolutionary computation algorithms, such as genetics algorithms and di erential evolution algorithms. For this, it will be detailed the methodology used for the construction of the chromosomes and the tness function. At least, these models will be compared with an interval predictor method, the quantile regression.
Description
Keywords
Energía solar, Modelos de predicción, Inteligencia artificial, Redes de neuronales, Algoritmos, Computación evolutiva