Aware Spatio-Temporal Graph Neural Network for Turbine-Level Wind Power Prediction

Accurate wind power forecasting at the turbine level is pivotal for optimizing grid integration, strengthening energy management strategies, and maximizing operational efficiency in the large-scale wind farms. Complex spatial interconnections between turbines and dynamic wake interactions influence power generation and substantially not included in conventional forecasting approaches. We propose a Wake-Aware SpatioTemporal Graph Neural Network (WAST-GNN) framework for turbine-level wind power prediction. In the proposed model, wind turbines are modelled as nodes of a graph and the wakes and spatial proximity as connectedness of the graph.

Govind Gupta¹, Lalit Agarwal^2*

¹Department of Electrical and Electronics Engineering, Maharaja Agrasen Institute of technology, Delhi, India
²Department of Electronics and Communication Engineering, Maharaja Agrasen Institute of technology, Delhi, India
* Corresponding Author. E-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract: Accurate wind power forecasting at the turbine level is pivotal for optimizing grid integration, strengthening energy management strategies, and maximizing operational efficiency in the large-scale wind farms. Complex spatial interconnections between turbines and dynamic wake interactions influence power generation and substantially not included in conventional forecasting approaches. We propose a Wake-Aware SpatioTemporal Graph Neural Network (WAST-GNN) framework for turbine-level wind power prediction. In the proposed model, wind turbines are modelled as nodes of a graph and the wakes and spatial proximity as connectedness of the graph. A sequential input modelling is utilized to incorporate temporal dependencies of past wind characteristics and power generation data to enable the collaborative learning of temporal and spatial patterns. The framework is tested with synthetic Supervisory Control and Data Acquisition (SCADA) data and wind farm layouts. The proposed methodology is compared with baseline models namely traditional Graph Convolutional Networks (GCN), GraphSAGE, and multilayer perceptrons (MLP). Experimental results demonstrate that the proposed WAST-GNN provides lower root mean square error (RMSE) of 0.227 and an improved coefficient of determination 𝑅 2of 0.861 superior than the standard graph-based approaches. Results illustrate the effectiveness of using the wake-aware spatial relations and the temporal dynamics for improving operational efficiency using Graph Neural Networks (GNN). 

Keywords: Wind power prediction, Graph Neural Network, Wake effects, Spatio-temporal modeling, Graph Attention Network

REFRENCES

[1]. Clarke, “Wind energy progress and potential,” Energy Policy, vol. 19, no. 8, pp. 742–755, 1991, doi: 10.1016/0301-4215(91)90044-O.
[2]. X. Wang, C. Cai, S.-G. Cai, T. Wang, Z. Wang, J. Song, X. Rong, and Q. Li, “A review of aerodynamic and wake characteristics of floating offshore wind turbines,” Renewable and Sustainable Energy Reviews, vol. 175, p. 113144, 2023, doi: 10.1016/j.rser.2022.113144.
[3]. A. Keramatfar, M. Rafiee, and H. Amirkhani, “Graph neural networks: A bibliometrics overview,” Machine Learning with Applications, vol. 10, p. 100401, 2022, doi: 10.1016/j.mlwa.2022.100401.
[4]. E. A. Tuncar, Ş. Sağlam, and B. Oral, “A review of short-term wind power generation forecasting methods in recent technological trends,” Energy Reports, vol. 12, pp. 197–209, 2024, doi: 10.1016/j.egyr.2024.06.006.
[5]. R. Meka, A. Alaeddini, and K. Bhaganagar, “A robust deep learning framework for short-term wind power forecast of a full-scale wind farm using atmospheric variables,” Energy, vol. 221, p. 119759, 2021, doi: 10.1016/j.energy.2021.119759.
[6]. R. Shrivastava, I. S. Iyer, and A. Batra, “Short term wind prediction with an ARIMA model,” Wind Energy and Engineering Research, vol. 4, p. 100020, 2025, doi: 10.1016/j.weer.2025.100020.
[7]. E. A. Tuncar, Ş. Sağlam, and B. Oral, “A review of short-term wind power generation forecasting methods in recent technological trends,” Energy Reports, vol. 12, pp. 197–209, 2024, doi: 10.1016/j.egyr.2024.06.006.
[8]. L. Agarwal, B. Jaint, and A. K. Mandpura, “Reducing overfitting in deep learning intrusion detection for power systems with CTGAN,” Chaos, Solitons & Fractals, vol. 188, p. 115603, 2024, doi: 10.1016/j.chaos.2024.115603.
[9]. S. Li, A. Robert, A. A. Faisal, and M. D. Piggott, “Learning to optimise wind farms with graph transformers,” Applied Energy, vol. 359, p. 122758, 2024, doi: 10.1016/j.apenergy.2024.122758
[10].Wu, S., Xiong, Y., Liang, H. et al. D2-GCN: a graph convolutional network with dynamic disentanglement for node classification. Front. Comput. Sci. 19, 191305 (2025). https://doi.org/10.1007/s11704-023-3339-7
[11].S. Saidane, F. Telch, K. Shahin, and F. Granelli, “Deep GraphSAGE enhancements for intrusion detection: Analyzing attention mechanisms and GCN integration,” Journal of Information Security and Applications, vol. 90, p. 104013, 2025, doi: 10.1016/j.jisa.2025.104013
[12].A. Verdone, S. Scardapane, and M. Panella, “Explainable spatio-temporal graph neural networks for multisite photovoltaic energy production,” Applied Energy, vol. 353, pt. B, p. 122151, 2024, doi: 10.1016/j.apenergy.2023.122151
[13].Agarwal, L., Manisha, Arya, Y., Nagpal, N., Majumdar, R., & Gupta, V. (2025). A case study on the feasibility and optimization of wind farm deployment in India. Energy Sources, Part A: Recovery, Utilization, and Environmental Effects, 47(2). https://doi.org/10.1080/15567036.2025.2537739
[14].A. Naimo, “A novel approach to generate synthetic wind data,” Procedia - Social and Behavioral Sciences, vol. 108, pp. 187–196, 2014, doi: 10.1016/j.sbspro.2013.12.830.
[15].N. Zhu, Y. Wang, K. Yuan, J. Yan, Y. Li, and K. Zhang, “GGNet: A novel graph structure for power forecasting in renewable power plants considering temporal lead-lag correlations,” Applied Energy, vol. 364, p. 123194, 2024, doi: 10.1016/j.apenergy.2024.123194..
[16].A. G. Gonzalez-Rodriguez, J. Serrano-Gonzalez, M. Burgos-Payan, and J. Riquelme-Santos, “Complete dataset to be used as a workbench to evaluate the profitability of an offshore wind farm,” Data in Brief, vol. 43, p. 108396, 2022, doi: 10.1016/j.dib.2022.108396.
[17].L. Agarwal, B. Jaint, and A. K. Mandpura, “Hybrid AI framework for detecting cyberattacks and predicting cascading failures in power systems,” Sustainable Computing: Informatics and Systems, vol. 48, p. 101222, 2025, doi: 10.1016/j.suscom.2025.101222.