The data-based adaptive graph learning network for analysis and prediction of offshore wind speed. (15th March 2023)

• Record Type: Journal Article
• Authors: Ren, Yuting; Li, Zhuolin; Xu, Lingyu; Yu, Jie
• Is Part Of: Energy. Volume 267(2023)
• Publication Date: 2023-03-15
• Subjects: Intelligent prediction of offshore wind -- Spatio-temporal dependence -- Graph neural network -- Adaptive graph learning; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.126590
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Offshore wind power plays an important role in the economy because of its abundant resources and great potential. Therefore, predicting offshore wind power significantly affects the intelligent management of power generation. However, tackling such forecasting task usually meet huge challenges due to the complex-temporal dependence on offshore wind data. Recently, deep learning approaches have successfully demonstrated their ability in modeling time series data. However, they often have significant limitations for failing to explore dynamic spatio-temporal dependencies between signals. In this paper, we propose a new framework DAGLN, which performs spatial dependency modelling through data-driven graph construction and graph learning, breaking through the limitations of predefined graph structures to obtain high-dimensional spatial features and capturing temporal information from them based on GRU structure. The model can play a powerful role in mining spatio-temporal correlations in multi-node and multi-step wind speed data prediction. Extensive experiments on selected nodes and data in the China Sea show the developed approach can outperform state-of-art models in multi-node wind speed prediction. Highlights: A graph convolutional model for multi-node offshore wind speed prediction is proposed. The model can adaptively explore the variable spatial dependence between offshore wind nodes based on data. Personalized node parameter patterns are captured. Temporal information is obtained based on GRU. The model can predict wind speed accurately.