Publication: Accidentes en turbinas eólicas: Un estudio de minería de datos
Loading...
Official URL
Full text at PDC
Publication Date
2022
Authors
Advisors (or tutors)
Pandit, Ravi
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
En la lucha contra el cambio climático la energía eólica está jugando un papel crucial en la sustitución de energías fósiles. Como aún se trata de una tecnología en fase de expansión y desarrollo los accidentes son eventos que suceden de manera recurrente y en este trabajo nos planteamos realizar un análisis de estos. Este trabajo presenta los resultados obtenidos tras el estudio de 273 accidentes en aerogeneradores por todo el mundo.
Para ello procederemos a un análisis estadístico con la finalidad de ver la cuales son los elementos más relevantes a tener en cuenta en el accidente y como se relacionan estos con la posibilidad de que el accidente resulte en muertes o lesiones. Para este cometido emplearemos diversas herramientas estadísticas para abordar el análisis desde diversos enfoques. También procederemos a emplear métodos de selección y ranking de atributos, así como un análisis exploratorio de datos. La finalidad de estos procesos es ampliar el análisis a un estudio no solo de la relación entre los atributos, sino también a un estudio de los valores de esos atributos.
Finalmente, crearemos modelos predictivos utilizando varios algoritmos de clasificación para, en futuros casos, poder prever y evitar accidentes basándonos en los atributos estudiados. Los algoritmos empleados son de dos tipos, de aprendizaje supervisado como puede ser el algoritmo de ’random forest’ o ’k-nearest neighbor’ y algoritmos de aprendizaje no supervisados como pueden los algoritmos de ’k-means’ o ’affinity propagation’. En este segundo grupo también se estudiarán un par de arquitecturas de redes neuronales.
In the fight against climate change, wind energy is playing a crucial role in the substitution of fossil fuels. As it is still a technology in a phase of expansion and development, accidents are recurrent events and in this paper we propose to analyze them. This paper presents the results obtained after the study of 273 wind turbine accidents around the world. We will proceed to a statistical analysis in order to see which are the most relevant elements to take into account in the accident and how they are related to the possibility of the accident resulting in deaths or injuries. For this purpose we will use various statistical tools to approach the analysis from different approaches. We will also perform attribute selection and ranking procedures, as well as exploratory data analysis. The purpose of these processes is to extend the analysis to a study not only of the relationship between attributes, but also to a study of the values of those attributes. Finally, we will create predictive models using various classification algorithms in order to, in future cases, be able to predict and avoid accidents based on the attributes studied. The algorithms used are of two types, supervised learning algorithms such as the ’random forest’ or ’k-nearest neighbor’ algorithm and unsupervised learning algorithms such as the ’k-means’ or ’affinity propagation’ algorithms. In this second group, a couple of neural network architectures will also be studied.
In the fight against climate change, wind energy is playing a crucial role in the substitution of fossil fuels. As it is still a technology in a phase of expansion and development, accidents are recurrent events and in this paper we propose to analyze them. This paper presents the results obtained after the study of 273 wind turbine accidents around the world. We will proceed to a statistical analysis in order to see which are the most relevant elements to take into account in the accident and how they are related to the possibility of the accident resulting in deaths or injuries. For this purpose we will use various statistical tools to approach the analysis from different approaches. We will also perform attribute selection and ranking procedures, as well as exploratory data analysis. The purpose of these processes is to extend the analysis to a study not only of the relationship between attributes, but also to a study of the values of those attributes. Finally, we will create predictive models using various classification algorithms in order to, in future cases, be able to predict and avoid accidents based on the attributes studied. The algorithms used are of two types, supervised learning algorithms such as the ’random forest’ or ’k-nearest neighbor’ algorithm and unsupervised learning algorithms such as the ’k-means’ or ’affinity propagation’ algorithms. In this second group, a couple of neural network architectures will also be studied.
Description
Trabajo de Fin de Grado en Ingeniería de Software, Facultad de Informática UCM, Departamento de de Arquitectura de Computadores y Automática, Curso 2021/2022.