.png){kind=icon}
.png){kind=icon}
Return
Wind Power Interval Prediction Based on Multidecomposition and Kernel Density Pinball
.png){kind=icon}
PRE
AI
DOI：10.1109/tii.2026.3689642.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Wind power forecasting plays a crucial role in optimizing the management of new energy resources. However, the strong nonstationarity and complex temporal patterns of wind power generation hinder existing methods from accurately quantifying forecasting uncertainty. To address this issue, this article proposes a decomposition-enhanced probabilistic forecasting framework, termed ProbAttention MoveAvg gated recurrent unit (PMGRU), for wind power interval prediction, which improves uncertainty quantification through the joint design of sequence decomposition, temporal dependence modeling, and density-aware interval learning. In this framework, variational mode decomposition is used to reduce sequence nonstationarity, while a GRU network with Top-$K$ sparse self-attention and moving-average decomposition is constructed to capture temporal dependencies and trend information. In addition, a kernel-density-weighted pinball loss is developed to improve the tradeoff between interval reliability and sharpness, and a simulated-annealing-based particle-swarm-inspired optimization strategy is employed to adaptively tune key decomposition parameters. Experimental results show that the proposed method achieves a favorable overall performance against benchmark models. Moreover, strictly causal online rolling-window deployment reveals dataset-dependent performance degradation: endpoint effects are observed in both datasets, while stronger data variability further aggravates degradation by increasing cross-window mode inconsistency.
Keywords:
Decomposition
interval prediction
kernel -density-weighted pinball loss (KDE-Pinball)
sparse attention
Journal
IF:
9.9
Papers:
8.3K
Citations:
6.0W