Multistation Wind Speed Forecasting Based on Dynamic Spatiotemporal Graph Convolutional Networks

Wind speed forecasting is significant in practical applications such as energy dispatch and meteorological early warning systems. However, spatiotemporal correlations of wind speed are dynamically influenced by weather conditions, seasonal variations, and diurnal fluctuations, resulting in constantl...

Full description

Saved in:
Bibliographic Details
Main Authors: Jianhong Gan, Runqing Kang, Xun Deng, Chentao Mao, Zhibin Li, Peiyang Wei, Chunjiang Wu, Tongli He
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Subjects:
Online Access:https://ieeexplore.ieee.org/document/11078154/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Wind speed forecasting is significant in practical applications such as energy dispatch and meteorological early warning systems. However, spatiotemporal correlations of wind speed are dynamically influenced by weather conditions, seasonal variations, and diurnal fluctuations, resulting in constantly changing spatiotemporal patterns. Meanwhile, wind speed data inherently include short-term high-frequency fluctuations and long-term low-frequency trends, which traditional methods struggle to adapt to and accurately capture multiscale features. This article proposes a dynamic spatiotemporal graph convolutional network (DSTGFP) model for multistation wind speed prediction to address this challenge. First, the model constructs an adaptive dynamic adjacency matrix by integrating geographical locations among stations, dynamic time warping, and mutual information, facilitating dynamic graph modeling. Next, we introduce a novel spatiotemporal feature extraction framework, which employs residual graph convolutional networks combined with a multihead attention mechanism to extract spatial features. We simultaneously integrate temporal- and frequency-domain convolutions to capture multiscale temporal-frequency features. Finally, the particle swarm optimization algorithm is used for hyperparameter optimization to improve the prediction accuracy. Experimental results demonstrate that the DSTGFP model achieves reductions of 24.66% in the mean absolute error and 25.47% in the root-mean-square error compared to existing deep learning methods, highlighting its superior predictive performance.
ISSN:1939-1404
2151-1535