Deep learning-based forecasting of the automatic Frequency Reserve Restoration band price in the Iberian electricity market
comunitat-uji-handle:10234/9
comunitat-uji-handle2:10234/7034
comunitat-uji-handle3:10234/8619
comunitat-uji-handle4:
INVESTIGACIONMetadatos
Título
Deep learning-based forecasting of the automatic Frequency Reserve Restoration band price in the Iberian electricity marketFecha de publicación
2023Editor
ElsevierISSN
2352-4677Cita bibliográfica
CARDO-MIOTA, Javier; PÉREZ, Emilio; BELTRAN, Hector. Deep learning-based forecasting of the automatic Frequency Reserve Restoration band price in the Iberian electricity market. Sustainable Energy, Grids and Networks, 2023, vol. 35, p. 101110Tipo de documento
info:eu-repo/semantics/articleVersión de la editorial
https://www.sciencedirect.com/science/article/pii/S2352467723001182Versión
info:eu-repo/semantics/publishedVersionPalabras clave / Materias
Resumen
The replacement of conventional and dispatchable generation technologies by intermittent renewable
energy sources increases the need for ancillary services. New agents, such as batteries, may join
frequency regulation ... [+]
The replacement of conventional and dispatchable generation technologies by intermittent renewable
energy sources increases the need for ancillary services. New agents, such as batteries, may join
frequency regulation markets but they require accurate information about future market prices and
service demand trends in order to make their participation profitable. This paper proposes and analyzes
the accuracy of various deep learning-based models to estimate the secondary reserve marginal band
price in the automatic frequency restoration reserves service of the Iberian electricity market. First, a
correlation analysis allows determining various subsets of market variables used as model inputs. These
subsets include some highly correlated variables together with different combinations of others whose
influenced is analyzed. Next, three different neural network techniques are considered: feedforward,
convolutional and recurrent networks. For each of them, a random search is performed to obtain the
best set of hyperparameters. The analysis of the results shows how the LSTM model returns the best
performance metrics (63.22 % of mean absolute scaled error), clearly improving the state-of-the-art in
the domain. [-]
Publicado en
Sustainable Energy, Grids and Networks, 2023, vol. 35, p. 101110Entidad financiadora
European Commission | European Regional Development Fund | Agencia Estatal de Investigación | Universitat Jaume I
Código del proyecto o subvención
PID2021-125634OB-I00 | TED2021-130120B-C22 | PREDOC/2020/35 | UJI-B2021-35
Derechos de acceso
info:eu-repo/semantics/openAccess
Aparece en las colecciones
- ESID_Articles [483]