Keywords
IEM-E (KL0-elecrica), Machine Learning, clasificación, cargas domésticas, desagregación del consumo., Machine Learning, classification, domestic loads, consumption disaggregation.
Adaptación de cargas domésticas por el método de Machine Learning
Abstract
La inestabilidad del actual modelo energético lleva a numerosas y diversas instituciones a comprometerse a orientar al conjunto de las economías hacia un modelo más sostenible y respetuoso con el medio ambiente. Uno de los elementos clave para asegurar una transición energética es la electrificación de la economía.
En la economía actual, el sector de la edificación, que engloba los subsectores residencial y de servicios, representa un 20% de la energía final en España. Por tanto, la electrificación del subsector residencial puede ser especialmente interesante.
Por otra parte, conviene recordar que el Aprendizaje Automático, o Machine Learning en inglés, es una nueva tecnología que ha adquirido una gran popularidad y que pretende desarrollar la capacidad de aprendizaje de las máquinas.
Pues bien, el presente Proyecto de Fin de Grado pretende aplicar los métodos de Machine Learning a la electrificación del sector residencial, mediante un estudio de las cargas domésticas.
Numerosos estudios demuestran que gran parte del gasto eléctrico en las viviendas se debe a la incapacidad de los usuarios de estimar correctamente su consumo. Un desglose de la factura eléctrica permitiría la optimización del uso energético de los usuarios.
El método de Machine Learning seleccionado para cumplir con el objetivo de este proyecto es el clasificador de los k vecinos más cercanos. Se trata de un algoritmo de clasificación, que predice las etiquetas de objetos de clase desconocida.
El concepto de carga se puede definir de distintas maneras, por lo que el reconocimiento de cargas domésticas se puede plantear desde varios puntos de vista. Un primer planteamiento sería reducir la carga a una señal temporal que defina una determinada magnitud eléctrica que le sea característica. Un segundo planteamiento permitiría poner en práctica la teoría de la desagregación de consumos mediante el uso de una base de datos de libre acceso.
The instability of the current energetic model has led numerous institutions to commit, by setting objectives and developing measures to achieve them, to guide economies across the world to adopt a more sustainable model. In today’s economy, every sector makes use of some form of energy, being the transport and construction sector its heaviest users. The construction sector, that encompasses the residential and services subsectors, represents 20% of the final energy consumption in Spain. Subsequently, the electrification of the residential subsector can be of special interest to reach the stated objectives defined by the European Comission. Likewise, it should be emphazised that the improvement rate of new technologies has been exponential. A technology that has become popular over the last few years even though it is still in its earlier development phase, is the so-called Machine Learning, that pretends to develop the learning capabilities of machines. That being said, this Thesis Paper tries to apply Machine Learning methods to the electrification of the residential sector, through the study of domestic loads. Several studies have proven that most of the residential electricity consumption is due to the inability of the users to identify consumption. A breakdown of the power bill would increase energy efficiency. The selected Machine Learning method is the k-Nearest Neighbours classifier. It is a classification algorithm, that predicts labels of unknown objects. It creates, from a series of individual examples, a generalized model that is useful to identify future objects. The concept of the load can be defined in different ways, so the recognition of identical loads can be done from different perspectives. A first approach would be to reduce the load to a temporary signal that defines the electricity magnitude as familiar. A second approach would enable the implementation of the consumption dissagregation theory.
The instability of the current energetic model has led numerous institutions to commit, by setting objectives and developing measures to achieve them, to guide economies across the world to adopt a more sustainable model. In today’s economy, every sector makes use of some form of energy, being the transport and construction sector its heaviest users. The construction sector, that encompasses the residential and services subsectors, represents 20% of the final energy consumption in Spain. Subsequently, the electrification of the residential subsector can be of special interest to reach the stated objectives defined by the European Comission. Likewise, it should be emphazised that the improvement rate of new technologies has been exponential. A technology that has become popular over the last few years even though it is still in its earlier development phase, is the so-called Machine Learning, that pretends to develop the learning capabilities of machines. That being said, this Thesis Paper tries to apply Machine Learning methods to the electrification of the residential sector, through the study of domestic loads. Several studies have proven that most of the residential electricity consumption is due to the inability of the users to identify consumption. A breakdown of the power bill would increase energy efficiency. The selected Machine Learning method is the k-Nearest Neighbours classifier. It is a classification algorithm, that predicts labels of unknown objects. It creates, from a series of individual examples, a generalized model that is useful to identify future objects. The concept of the load can be defined in different ways, so the recognition of identical loads can be done from different perspectives. A first approach would be to reduce the load to a temporary signal that defines the electricity magnitude as familiar. A second approach would enable the implementation of the consumption dissagregation theory.
Información detallada
Materias, derechos, colecciones e identificadores