Wind Power Forecasting Based on Multi-Graph Neural Networks Considering External Disturbances
Abstract
:1. Introduction
- A two-graph convolutional network is proposed for extracting spatial features from different farms. Specifically, two different adjacency matrices are constructed to describe similarity among nodes, indicating that more similar nodes have more similar characteristics.
- To capture the temporal features of a farm, a Long Short-Term Memory (LSTM) network is designed, which uses an attention mechanism to distinguish the importance of different temporal features.
- A combinational mechanism is designed to further incorporate the EIF into wind power prediction. First, the EIF is eliminated from the historical wind power to represent inherent wind power patterns. Next, the wind power without the EIF is fed into LSTM-Attention. Finally, to better account for future disturbances, the final prediction is derived by integrating the EIF of the prediction time interval with the output of LSTM-Attention. A comprehensive series of experiments is carried out to confirm the efficacy of the EIF modeling approach and the combinatorial mechanism.
2. Related Work
3. Methodology
3.1. Spatial Features and Multi-Graph Convolution Networks
- Neighborhood graph : The neighborhood graph performs information aggregation from neighboring nodes and utilizes their physical proximity to assess their correlation. Generally, a shorter distance between two nodes indicates a stronger correlation between them. Therefore, the elements in the adjacent matrix are defined using the Euclidean distance between nodes:
- Spatial power correlation graph : The Pearson coefficient is employed to examine the spatial correlation between the power across various nodes. Therefore, it is utilized to establish the adjacent matrix as
3.2. Modeling External Interference Factors
3.3. LSTM Networks and Attention Mechanisms
3.4. Parameter Learning
4. Experiment
4.1. Dataset Description
4.2. Data Process
4.3. Hyperparameter Optimization
4.4. Evaluation Indicators
5. Results and Analysis
- Comparison with baselines. We begin by comparing the general predictive performance between GCN-EIF and baselines; next, we analyze our model’s computational complexity and real-time performance; third, we examine the model performance on various days, including workdays, weekends, and holidays.
- Computational efficiency analysis. We analyze the theoretical complexity of GCN-EIF and perform comprehensive empirical evaluations of runtime performance, processing latency, and memory scaling in different operational scenarios to verify the suitability of the model for real-time applications.
- Comparison with variants of GCN-EIF. First, we validate the efficacy of the GCN-EIF modeling approach; subsequently, we investigate the influence of the elimination and combination approaches; lastly, we assess the impact of multiple context factors.
- Impact of hyperparameters. We also explore the influence of critical hyperparameters in GCN-EIF, including the number of LSTM layers, historical time intervals, and selection of similar nodes for graph construction.
5.1. Comparison with Baselines
5.2. Comparison with Variants of GCN-EIF
5.3. Ablation Study on Attention Mechanisms
5.4. Impact of Hyperparameters
6. Conclusions
Author Contributions
Funding
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Bazionis, I.K.; Georgilakis, P.S. Review of deterministic and probabilistic wind power forecasting: Models, methods, and future research. Electricity 2021, 2, 13–47. [Google Scholar] [CrossRef]
- Wang, Y.; Zou, R.; Liu, F.; Zhang, L.; Liu, Q. A review of wind speed and wind power forecasting with deep neural networks. Appl. Energy 2021, 304, 117766. [Google Scholar] [CrossRef]
- Biswas, A.K.; Ahmed, S.I.; Bankefa, T.; Ranganathan, P.; Salehfar, H. Performance analysis of short and mid-term wind power prediction using ARIMA and hybrid models. In Proceedings of the 2021 IEEE Power and Energy Conference at Illinois (PECI), Urbana, IL, USA, 1–2 April 2021; pp. 1–7. [Google Scholar]
- Deng, Y.-C.; Tang, X.-H.; Zhou, Z.-Y.; Yang, Y.; Niu, F. Application of machine learning algorithms in wind power: A review. Energy Sources Part A Recover. Util. Environ. Eff. 2021, 47, 4451–4471. [Google Scholar] [CrossRef]
- Chen, J.; Zhu, Q.; Li, H.; Zhu, L.; Shi, D.; Li, Y.; Duan, X.; Liu, Y. Learning Heterogeneous Features Jointly: A Deep End-to-End Framework for Multi-Step Short-Term Wind Power Prediction. IEEE Trans. Sustain. Energy 2020, 11, 1761–1772. [Google Scholar] [CrossRef]
- Sun, Z.; Zhao, M. Short-Term Wind Power Forecasting Based on VMD Decomposition, ConvLSTM Networks and Error Analysis. IEEE Access 2020, 8, 134422–134434. [Google Scholar] [CrossRef]
- Saint-Drenan, Y.-M.; Besseau, R.; Jansen, M.; Staffell, I.; Troccoli, A.; Dubus, L.; Schmidt, J.; Gruber, K.; Simões, S.G.; Heier, S. A parametric model for wind turbine power curves incorporating environmental conditions. Renew. Energy 2020, 157, 754–768. [Google Scholar] [CrossRef]
- Ozpineci, B. AI/ML and Cybersecurity in Power Electronics. IEEE Power Electron. Mag. 2022, 9, 38–40. [Google Scholar] [CrossRef]
- Zhao, S.; Wang, H. Enabling Data-Driven Condition Monitoring of Power Electronic Systems With Artificial Intelligence: Concepts, Tools, and Developments. IEEE Power Electron. Mag. 2021, 8, 18–27. [Google Scholar] [CrossRef]
- Gupta, A.; Sharma, K.C.; Vijayvargia, A.; Bhakar, R. Very short term wind power prediction using hybrid univariate ARIMA-GARCH model. In Proceedings of the 2019 8th International Conference on Power Systems (ICPS), Jaipur, India, 20–22 December 2019; pp. 1–6. [Google Scholar]
- Xu, T.; Du, Y.; Li, Y.; Zhu, M.; He, Z. Interval prediction method for wind power based on VMD-ELM/ARIMA-ADKDE. IEEE Access 2022, 10, 72590–72602. [Google Scholar] [CrossRef]
- Alkesaiberi, A.; Harrou, F.; Sun, Y. Efficient wind power prediction using machine learning methods: A comparative study. Energies 2022, 15, 2327. [Google Scholar] [CrossRef]
- Zhang, Y.; Sun, H.; Guo, Y. Wind power prediction based on PSO-SVR and grey combination model. IEEE Access 2019, 7, 136254–136267. [Google Scholar] [CrossRef]
- Shi, Z.; Liang, H.; Dinavahi, V. Direct Interval Forecast of Uncertain Wind Power Based on Recurrent Neural Networks. IEEE Trans. Sustain. Energy 2018, 9, 1177–1187. [Google Scholar] [CrossRef]
- Niu, Z.; Yu, Z.; Tang, W.; Wu, Q.; Reformat, M. Wind power forecasting using attention-based gated recurrent unit network. Energy 2020, 196, 117081. [Google Scholar] [CrossRef]
- Pan, C.; Wen, S.; Zhu, M.; Ye, H.; Ma, J.; Jiang, S. Hedge Backpropagation Based Online LSTM Architecture for Ultra-Short-Term Wind Power Forecasting. IEEE Trans. Power Syst. 2024, 39, 4179–4192. [Google Scholar] [CrossRef]
- Liu, M.; Qin, H.; Cao, R.; Deng, S. Short-Term Load Forecasting Based on Improved TCN and DenseNet. IEEE Access 2022, 10, 115945–115957. [Google Scholar] [CrossRef]
- Song, Y.; Tang, D.; Yu, J.; Yu, Z.; Li, X. Short-Term Forecasting Based on Graph Convolution Networks and Multiresolution Convolution Neural Networks for Wind Power. IEEE Trans. Ind. Inform. 2023, 19, 1691–1702. [Google Scholar] [CrossRef]
- Liao, W.; Wang, S.; Bak-Jensen, B.; Pillai, J.R.; Yang, Z.; Liu, K. Ultra-short-term Interval Prediction of Wind Power Based on Graph Neural Network and Improved Bootstrap Technique. J. Mod. Power Syst. Clean Energy 2023, 11, 1100–1114. [Google Scholar] [CrossRef]
- Li, Z.; Ye, L.; Song, X.; Luo, Y.; Pei, M.; Wang, K.; Yu, Y.; Tang, Y. Heterogeneous Spatiotemporal Graph Convolution Network for Multi-Modal Wind-PV Power Collaborative Prediction. IEEE Trans. Power Syst. 2024, 39, 5591–5608. [Google Scholar] [CrossRef]
- Tang, J.; Liu, Z.; Hu, J. Spatial-Temporal Wind Power Probabilistic Forecasting Based on Time-Aware Graph Convolutional Network. IEEE Trans. Sustain. Energy 2024, 15, 1946–1956. [Google Scholar] [CrossRef]
- Pryor, S.C.; Barthelmie, R.J.; Bukovsky, M.S.; Leung, L.R.; Sakaguchi, K. Climate change impacts on wind power generation. Nat. Rev. Earth Environ. 2020, 1, 627–643. [Google Scholar] [CrossRef]
Timestamp | 1 | 2 | … | 40 |
---|---|---|---|---|
2022/1/1 0:00 | 6 | 5 | 1.568 | |
2022/1/1 1:00 | 2 | 10 | … | 2.491 |
… | … | |||
2023/8/31 23:00 | 48.23 | 69.9 | … | 29.62 |
Feature | Sample |
---|---|
Longitude | 32.6° |
Latitude | 121.1° |
Timestamp | 2022/1/1 0:00 |
Pressure | 103,138.92 Pa |
Wind Speed | 8.16 m/s |
Wind Direction | 29.34 |
Humidity | 39.96% |
Temperature | 279.69 K |
Algorithm | RMSE | MAE | R2 |
---|---|---|---|
ARIMA | 25.55 | 15.68 | 0.821 |
SVR | 22.25 | 14.87 | 0.833 |
RNN | 19.56 | 13.55 | 0.854 |
LSTM | 18.45 | 12.29 | 0.862 |
GRU | 18.34 | 12.31 | 0.865 |
CNN-LSTM | 17.69 | 10.39 | 0.876 |
CNN-LSTM-Attention | 17.25 | 9.58 | 0.882 |
STGCN | 15.69 | 8.26 | 0.897 |
GCN-EIF | 12.71 | 7.84 | 0.935 |
Algorithms | Workdays | Weekends | All Days |
---|---|---|---|
GRU | 19.39 | 18.67 | 18.34 |
CNN-LSTM | 17.38 | 18.26 | 17.69 |
CNN-LSTM-Attention | 16.55 | 17.32 | 17.25 |
STGCN | 14.63 | 15.21 | 15.69 |
GCN-EIF | 12.96 | 13.34 | 12.71 |
Algorithm | Training Time (min/ep) | Inference Time (ms/samp) | GPU Memory (MB) | Model Size (MB) |
---|---|---|---|---|
ARIMA | 0.5 | 2.3 | N/A | 0.8 |
SVR | 3.2 | 1.8 | N/A | 4.5 |
LSTM | 2.5 | 3.5 | 128 | 12.4 |
CNN-LSTM | 3.7 | 5.2 | 246 | 18.7 |
STGCN | 4.1 | 7.6 | 312 | 23.5 |
GCN-EIF | 5.8 | 8.3 | 358 | 26.2 |
Algorithm | RMSE | MAE | R2 |
---|---|---|---|
14.69 | 8.62 | 0.915 | |
13.35 | 8.21 | 0.921 | |
+ | 12.71 | 7.84 | 0.935 |
Algorithm | RMSE | MAE | R2 |
---|---|---|---|
GCN | 13.68 | 8.37 | 0.921 |
GCN-LSTM | 13.12 | 8.21 | 0.928 |
GCN-LSTM-Attention | 13.04 | 8.15 | 0.931 |
GCN-EIF | 12.71 | 7.84 | 0.935 |
Method | RMSE | MAE | R2 |
---|---|---|---|
NoElimination + NoCombination | 14.25 | 8.66 | 0.915 |
Elimination + NoCombination | 13.25 | 8.26 | 0.925 |
NoElimination + Combination | 13.21 | 8.19 | 0.929 |
Elimination + Combination | 12.71 | 7.84 | 0.935 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2025 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Xu, X.; Luo, Z.; Feng, M. Wind Power Forecasting Based on Multi-Graph Neural Networks Considering External Disturbances. Energies 2025, 18, 2969. https://doi.org/10.3390/en18112969
Xu X, Luo Z, Feng M. Wind Power Forecasting Based on Multi-Graph Neural Networks Considering External Disturbances. Energies. 2025; 18(11):2969. https://doi.org/10.3390/en18112969
Chicago/Turabian StyleXu, Xiaoyin, Zhumei Luo, and Menglong Feng. 2025. "Wind Power Forecasting Based on Multi-Graph Neural Networks Considering External Disturbances" Energies 18, no. 11: 2969. https://doi.org/10.3390/en18112969
APA StyleXu, X., Luo, Z., & Feng, M. (2025). Wind Power Forecasting Based on Multi-Graph Neural Networks Considering External Disturbances. Energies, 18(11), 2969. https://doi.org/10.3390/en18112969