Probabilistic multi-step prediction of wind power based on online feature extraction and parallel output network structure

Cao, Chaojin, He, Yaoyao and Zhou, Yue ORCID: https://orcid.org/0000-0002-6698-4714 2026. Probabilistic multi-step prediction of wind power based on online feature extraction and parallel output network structure. Engineering Applications of Artificial Intelligence 181 (1) , 115298. 10.1016/j.engappai.2026.115298

Full text not available from this repository.

Abstract

Wind energy plays a pivotal role in alleviating energy shortages and mitigating global warming. The integration of large-scale wind power into the electrical grid necessitates accurate probabilistic forecasting to maintain system stability. However, producing dependable multi-step probabilistic forecasts is difficult because wind power is strongly influenced by external variables such as wind speed and direction. Since these variables change over time, the underlying data distribution also shifts, a challenge commonly referred to as concept drift. Conventional feature selection techniques are often inadequate for capturing such evolving input patterns under concept drift. To address this limitation, this paper proposes an online feature extraction (OFE) framework. The framework dynamically reselects the input features when concept drift occurs and generates wind power features using the newly selected feature set, which can avoid frequent model reconstruction caused by input inconsistencies. To fully utilize the extracted features, a hybrid model with a parallel output network structure is designed for multi-step probabilistic wind power forecasting. This model leverages the advantages of temporal convolutional networks (TCN) and bidirectional long short-term memory (BiLSTM) networks in time series learning. By generating all multi-step probabilistic forecasts in a single forward pass, the parallel output structure avoids the iterative inference required by autoregressive strategies, making it particularly suitable for real-time grid operation Experimental results demonstrate that the proposed approach improves probabilistic forecasting reliability by over 20% compared with benchmark models.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	0952-1976
Date of Acceptance:	31 May 2026
Last Modified:	18 Jun 2026 08:45
URI:	https://orca.cardiff.ac.uk/id/eprint/187614

Actions (repository staff only)

Edit Item