Load disaggregation based on supervised learning algorithms for energy consumption pattern recognition in residential buildings

[EN] Climate change and global warming are causing growing environmental concerns, prompting many countries to increase investments in renewable energies and green strategies. The high growth rate of intermittent energy sources in the energy supply systems is causing balancing challenges that need t...

Descripción completa

Detalles Bibliográficos
Autores: Curcio, Giovanni Maria, Di Somma, Marialaura, Bianco, Nicola, Montuori, Lina|||0000-0001-7574-7916, Alcázar-Ortega, Manuel|||0000-0001-5384-3931
Tipo de recurso: artículo
Fecha de publicación:2026
País:España
Institución:Universitat Politècnica de València (UPV)
Repositorio:RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia
Idioma:inglés
OAI Identifier:oai:dnet:riunet______::7f1ca0c12070b28738bbc2f4d2b99598
Acceso en línea:https://riunet.upv.es/handle/10251/233758
Access Level:acceso abierto
Palabra clave:Load disaggregation
Supervised Learning
Energy consumption
Energy management
Machine learning
Residential building
07.- Asegurar el acceso a energías asequibles, fiables, sostenibles y modernas para todos
Descripción
Sumario:[EN] Climate change and global warming are causing growing environmental concerns, prompting many countries to increase investments in renewable energies and green strategies. The high growth rate of intermittent energy sources in the energy supply systems is causing balancing challenges that need to be addressed. Thus, demandside flexibility is attracting interest as a viable solution to guarantee grid reliability. In this framework, the present study aims to identify consumption behaviors for both thermal and electrical loads in residential building sector to assess the flexibility potential that end-users can offer to the Distribution System Operator (DSO). Moreover, based on performance and flexibility capacity, the main energy appliances have been disaggregated by means of a Supervised Learning algorithms assessing consumption forecasting. The proposed model has been applied on a large dataset of residential customers. The results obtained show that the neural network is able to explain 91.7% of the variability of synthetic dataset (R2 = 0.917), with an average absolute error of 1,797 kW in the process of disaggregating the consumption of 14 household appliances belonging to 558 homes. Finally, this research study provides valuable guidelines to carry out short-term energy resource planning and alleviate the existing operational problems of the energy systems.