Cluster-based demand prediction: a hybrid approach with grey wolf optimizer and multilayer perceptron network
[EN] The increasing complexity of microgrids in residential and industrial settings demands adaptive demand forecasting methods. This study introduces a novel approach that clusters household consumption data to identify distinct energy profiles, enabling precise modeling. A multilayer perceptron ne...
| Autores: | , , , |
|---|---|
| Formato: | artículo |
| Fecha de publicación: | 2025 |
| País: | España |
| Recursos: | Universitat Politècnica de València (UPV) |
| Repositorio: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia |
| Idioma: | inglés |
| OAI Identifier: | oai:riunet.upv.es:10251/230553 |
| Acesso em linha: | https://riunet.upv.es/handle/10251/230553 |
| Access Level: | acceso abierto |
| Palavra-chave: | Demand forecasting Energy Clustering Grey wolf optimizer Random forest Artificial neural network 07.- Asegurar el acceso a energías asequibles, fiables, sostenibles y modernas para todos |
| Resumo: | [EN] The increasing complexity of microgrids in residential and industrial settings demands adaptive demand forecasting methods. This study introduces a novel approach that clusters household consumption data to identify distinct energy profiles, enabling precise modeling. A multilayer perceptron network, optimized using the grey wolf optimizer, is tailored to these profiles, dynamically capturing unique consumption behaviors. Also, a random forest model also predicts future demand profiles based on factors like date, day type, and temperature, ensuring accurate profile assignment for each forecasting period. The proposed model achieves high accuracy, with normalized root-mean-squared error values of 0.04 for Case A and 0.05 for Case B, and mean absolute errors of 0.03¿0.05 kW, respectively. The total predicted demand closely matches the real values (7.37¿kWh vs. 7.34¿kWh for Case A and 7.31¿kWh vs. 7.56¿kWh for Case B). Prediction times are low (1.7428 s for Case A, 0.8215 s for Case B), and model training takes approximately 54 min. Compared to existing methods from state of the art, the proposed solution offers lower errors. The proposed combination of clustering, machine learning, and metaheuristic optimization establishes a strong and efficient framework for microgrid demand forecasting. |
|---|