RT info:eu-repo/semantics/article
T1 Additive ensemble neural network with constrained weighted quantile loss for probabilistic electric-load forecasting
A1 López Martín, Manuel
A1 Sánchez Esguevillas, Antonio Javier
A1 Hernández Callejo, Luis
A1 Arribas Sánchez, Juan Ignacio
A1 Carro Martínez, Belén
K1 Short and medium-term electric-load forecasting
K1 Quantile forecasting
K1 Deep learning
K1 Machine learning
K1 Deep learning additive ensemble model
K1 33 Ciencias Tecnológicas
AB This work proposes a quantile regression neural network based on a novel constrainedweighted quantile loss (CWQLoss) and its application to probabilistic short and medium-termelectric-load forecasting of special interest for smart grids operations. The method allows any pointforecast neural network based on a multivariate multi-output regression model to be expanded tobecome a quantile regression model. CWQLoss extends the pinball loss to more than one quantileby creating a weighted average for all predictions in the forecast window and across all quantiles.The pinball loss for each quantile is evaluated separately. The proposed method imposes additionalconstraints on the quantile values and their associated weights. It is shown that these restrictions areimportant to have a stable and efficient model. Quantile weights are learned end-to-end by gradientdescent along with the network weights. The proposed model achieves two objectives: (a) produceprobabilistic (quantile and interval) forecasts with an associated probability for the predicted targetvalues. (b) generate point forecasts by adopting the forecast for the median (0.5 quantiles). Weprovide specific metrics for point and probabilistic forecasts to evaluate the results considering bothobjectives. A comprehensive comparison is performed between a selection of classic and advancedforecasting models with the proposed quantile forecasting model. We consider different scenarios forthe duration of the forecast window (1 h, 1-day, 1-week, and 1-month), with the proposed modelachieving the best results in almost all scenarios. Additionally, we show that the proposed methodobtains the best results when an additive ensemble neural network is used as the base model. Theexperimental results are drawn from real loads of a medium-sized city in Spain.
PB MDPI
SN 1424-8220
YR 2021
FD 2021
LK https://uvadoc.uva.es/handle/10324/54135
UL https://uvadoc.uva.es/handle/10324/54135
LA eng
NO Sensors, 2021, vol. 21, n. 9, p. 2979
NO Producción Científica
DS UVaDOC
RD 18-dic-2024