Accurate wind speed prediction using a Variational AutoEncoder integrated with a sequence-to-sequence-bidirectional long short-term memory-encoder-decoder architecture

Accurate wind speed prediction plays a pivotal role in optimizing wind energy systems and supporting sustainable environmental management. This study aims to addressing key challenges in wind speed prediction, including uncertainty, dimensionality reduction, and nonlinearity, by proposing a Variatio...

Ausführliche Beschreibung

Gespeichert in:

Bibliographische Detailangaben
Veröffentlicht in:	Engineering applications of artificial intelligence Jg. 163; S. 113090
Hauptverfasser:	Mehmood, Faisal, Yuan, Chulong, Gong, Jingliang, Li, Lixiao, Hu, Gang, Li, Chao, Abbas, Touqeer
Format:	Journal Article
Sprache:	Englisch
Veröffentlicht:	Elsevier Ltd 01.01.2026
Schlagworte:	Bidirectional long short-term memory Deep learning Variational AutoEncoder Wind energy Wind speed prediction Deep learning Wind speed prediction Wind energy Bidirectional long short-term memory Variational AutoEncoder
ISSN:	0952-1976
Online-Zugang:	Volltext
Tags:	Tag hinzufügen Keine Tags, Fügen Sie den ersten Tag hinzu!

Beschreibung
Zusammenfassung:	Accurate wind speed prediction plays a pivotal role in optimizing wind energy systems and supporting sustainable environmental management. This study aims to addressing key challenges in wind speed prediction, including uncertainty, dimensionality reduction, and nonlinearity, by proposing a Variational AutoEncoder (VAE) integrated with a Sequence-to-Sequence Bidirectional Long Short-Term Memory (Seq2Seq-BiLSTM) Encoder-Decoder model. The Variational AutoEncoder employs a Bayesian probabilistic framework to capture complex data distributions in a probabilistic latent space, enabling effective dimensionality reduction and robust handling of nonlinearities. The Sequence-to-Sequence-Bidirectional framework consists of two long short-term memory layers to capture temporal dependencies by processing time series data in both forward and backward directions for enhancing predictive accuracy. The encoder bidirectionally encodes input sequences, while the decoder generates step-by-step predictions, ensuring a comprehensive understanding of wind speed patterns over time. The proposed approach was validated using wind turbine data from Beijing, China, and benchmarked against several state-of-the-art deep learning models. Experimental results demonstrate that the proposed model outperforms competing methods, achieving a normalized mean square error of 0.0162 and an R2 of 0.9837. Feature extraction through nonlinear modeling revealed underlying temporal patterns, and visual analyses highlighted seasonal variations in wind, temperature, humidity, and power output. The data show stronger winds in cooler months and consistent distributions across weekdays and weekends, with stable yearly trends. The proposed framework offers a scalable and reliable solution for operational wind speed forecasting, with implications for energy dispatch planning and policy-making. To facilitate practical deployment, a Wind Speed Prediction App (WSP app) was developed based on the proposed model. All datasets and source code are available at: https://github.com/Faisal-code/Wind_Speed_Prediction.
ISSN:	0952-1976
DOI:	10.1016/j.engappai.2025.113090