Publication: Redes Adversarias Generativas para el Aumento de Datos Secuenciales 1D. Aplicación a turbinas eólicas
Loading...
Official URL
Full text at PDC
Publication Date
2022
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Las energías renovables son aquellas basadas en el uso de recursos naturales ilimitados. Se caracterizan por evitar la producción de gases de efecto invernadero (principal causa del cambio climático). Su desarrollo ha sido fuertemente impulsado en los últimos años.
Dentro de las energías renovables se encuentra la energía eólica, que aprovecha el efecto de las corrientes del aire. Dicha energía se recoge en parques eólicos, donde se encuentran los aerogeneradores, máquinas que transforman la fuerza del viento o energía cinética en energía eléctrica. Los parques eólicos están compuestos de turbinas eólicas que pueden dividirse en diversas categorías, centrándonos en el proyecto en las turbinas eólicas flotantes, situadas generalmente en zonas marinas.
El objetivo de este proyecto es implementar una red generativa adversaria o GAN (Generative Adversarial Network), que a partir de unos datos de entrada reales que contienen parámetros de configuración de las turbinas, sea entrenada para autogenerar datos sintéticos similares a los reales. Estos datos sintéticos son requeridos debido a la falta de datos reales en las turbinas marinas flotantes, facilitando poder estudiar problemas relacionados con desgaste o roturas de las turbinas eólicas y entrenar otros modelos de control.
Una GAN, o red generativa adversaria, es una clase de red neuronal compuesta por un algoritmo discriminativo o discriminador que, una vez correctamente entrenado, es capaz de distinguir entre imágenes reales y otras sintéticas. Estas imágenes sintéticas son desarrolladas por el algoritmo generativo o generador, otra red neuronal que llega a generar imágenes sintéticas con un parecido tan idéntico a las reales que hace confundir al discriminador.
Renewable energies are those based on the use of unlimited natural resources. They are characterised by avoiding the production of greenhouse gases (the main cause of climate change). Their development has been strongly boosted in recent years. Renewable energies include wind energy, which harnesses the effect of air currents. This energy is collected in wind farms, where wind turbines, machines that transform the force of the wind or kinetic energy into electrical energy, are located. Wind farms are made up of wind turbines that can be divided into different categories, with the project focusing on floating wind turbines, which are generally located in marine areas. The aim of this project is to implement a Generative Adversarial Network (GAN), which from real input data containing turbine configuration parameters, is trained to selfgenerate synthetic data similar to real data. This synthetic data is required due to the lack of real data on floating offshore turbines, making it easier to study problems related to wind turbine wear and tear and to train other control models. A GAN, or generative adversarial network, is a class of neural network composed of a discriminative algorithm or discriminator that, once properly trained, is able to distinguish between real and synthetic images. These synthetic images are evolved by the generative or generative algorithm, another neural network that manages to generate synthetic images that are so similar to the real ones that they confuse the discriminator.
Renewable energies are those based on the use of unlimited natural resources. They are characterised by avoiding the production of greenhouse gases (the main cause of climate change). Their development has been strongly boosted in recent years. Renewable energies include wind energy, which harnesses the effect of air currents. This energy is collected in wind farms, where wind turbines, machines that transform the force of the wind or kinetic energy into electrical energy, are located. Wind farms are made up of wind turbines that can be divided into different categories, with the project focusing on floating wind turbines, which are generally located in marine areas. The aim of this project is to implement a Generative Adversarial Network (GAN), which from real input data containing turbine configuration parameters, is trained to selfgenerate synthetic data similar to real data. This synthetic data is required due to the lack of real data on floating offshore turbines, making it easier to study problems related to wind turbine wear and tear and to train other control models. A GAN, or generative adversarial network, is a class of neural network composed of a discriminative algorithm or discriminator that, once properly trained, is able to distinguish between real and synthetic images. These synthetic images are evolved by the generative or generative algorithm, another neural network that manages to generate synthetic images that are so similar to the real ones that they confuse the discriminator.
Description
Trabajo de Fin de Grado en Ingeniería Informática, Facultad de Informática UCM, Departamento de Arquitectura de Computadores y Automática, Curso 2021/2022.