为提升超短期风电功率预测准确度、探索不同特征对功率预测模型的贡献,该文提出一种基于CEEMD-GA-BiGRU-Attention的风电超短期预测模型,并采用SHAP值分析方法分析各特征贡献。首先采用完全集成经验模态分解（CEEMD）方法分解功率数据,其次利用遗传算法（GA）优化带注意力机制的双向门控循环单元（BiGRU-attention）,对各模态分量动态建模,构建组合预测模型。将该模型与多种不同功率预测模型进行比较,结果表明其在RMSE、MAE、R2评价指标上具有优势,相较于BiGRU模型,R2提升10.08%,RMSE和MAE分别降低40.15%和42.38%。最后,采用SHAP值分析方法研究发现,对模型贡献较大的特征为历史实际功率和天气预报风速,前者在超短期预测初期占据主导地位,但重要性随时间步推进逐渐下降,后者的重要性则在预测中后期呈稳定上升趋势。
Abstract
To enhance the accuracy of ultra-short-term wind power prediction and investigate the contribution of diverse features to power prediction models, this paper introduces a wind power ultra-short-term prediction model grounded in CEEMD-GA-BiGRU-Attention. The model employs the SHAP value analysis method to assess the contribution of various features. Initially, the power data is decomposed using the Complete Ensemble Empirical Mode Decomposition （CEEMD） method. Subsequently, the Bidirectional Gated Recurrent Unit （BiGRU） model integrated with an attention mechanism is optimized using the Genetic Algorithm （GA） to dynamically model different modal components, thereby constructing a combined prediction model. When compared to various other power prediction models, this proposed model demonstrates superiority in terms of RMSE, MAE, and R2 evaluation metrics, with a notable 10.08% improvement in R2 compared to the BiGRU model, and a reduction of 40.15% and 42.38% in RMSE and MAE, respectively. Finally, through the application of the SHAP value analysis method, it is revealed that historical actual power and weather forecast wind speed are the features that significantly contribute to the model. Historical actual power dominates in the initial stages of ultra-short-term prediction, but its importance gradually diminishes over time, whereas the importance of weather forecast wind speed exhibits a stable upward trend in the middle to later stages.
关键词
风电功率预测 /
注意力机制 /
双向门控循环单元 /
可解释性
Key words
wind power prediction /
attention mechanism /
bidirectional gated recurrent unit /
interpretability
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 国家能源局. 2025年可再生能源并网运行情况[EB/OL].(2026-02-12)[2026-09-01]. https://www.nea.gov.cn/20260212/742b8c6a078347b0b39de676c05c5d58/c.html.
National Energy Administration. Description of renewable energy integration in2025 [EB/OL].(2026-02-12)[2026-09-01]. https://www.nea.gov.cn/20260212/742b8c6a078347b0b39de676c05c5d58/c.html.
[2] 国家能源局西北监管局. 国家能源局西北监管局关于印发《西北区域电力并网运行管理实施细则》《西北区域电力辅助服务管理实施细则》补充规则的通知[EB/OL]. [2025-04-17]. https://xbj.nea.gov.cn/dtyw/tzgg/202501/t20250102_274863.html
Northwest Regulatory Bureau of the National Energy Administration. Notice of the Northwest Regulatory Bureau of the National Energy Administration on the Issuance of Supplementary Rules to the "Implementation Rules for the Management of Grid-connected Operation of Electricity in the Northwest Region" and the "Implementation Rules for the Management of Auxiliary Services for Electricity in the Northwest Region"[EB/OL]. [2025-04-17]. https://xbj.nea.gov.cn/dtyw/tzgg/202501/t20250102_274863.html
[3] 唐杰, 李彬. 基于EMD-PSO-BiLSTM组合模型的短期风电功率预测[J]. 自动化应用, 2024, 65(5): 126-129.
Tang J, Li B.Short-term wind power prediction based on EMD-PSO-BiLSTM combined model[J]. Automation Application, 2024, 65(5): 126-129.
[4] 耿运涛. 基于EMD-PCA-LSTM的短期风电功率预测研究[J]. 船电技术, 2024, 44(11): 20-23.
Geng Y T.Short-term wind power prediction based on EMD-PCA-LSTM[J]. Marine Electric & Electronic Engineering, 2024, 44(11): 20-23.
[5] 鄢伟安, 付晟, 刘卫东, 等. 基于STL-EMD-SSA-LSTM模型的光伏中期输出功率预测研究[J]. 工业工程与管理, 2024, 29(6): 93-103.
Yan W A, Fu S, Liu W D, et al.Prediction of mid-term photovoltaic output power based on STL-EMD-SSA-LSTM model[J]. Industrial Engineering and Management, 2024, 29(6): 93-103.
[6] 王潇添, 李泽林, 詹莹, 等. 基于KOA驱动的VMD-CNN-BiGRU-Attention光伏功率组合预测模型研究[J]. 太阳能学报, 2026, 47(6): 676-688.
Wang X T, Li Z L, Zhan Y, et al.KOA-driven VMD-CNN-BiGRU-Attention models research for combined prediction of PV power output[J]. Acta Energiae Solaris Sinica, 2026, 47(6): 676-688.
[7] 潘鹏程, 张帅, 刘晖, 等. 基于CAWR-LSTM-TRF模型的超短期风电功率预测[J]. 太阳能学报, 2026, 47(3): 452-461.
Pan P C, Zhang S, Liu H, et al.Ultra-short-term wind power forecasting based on CAWR-LSTM-TRF model[J]. Acta Energiae Solaris Sinica, 2026, 47(3): 452-461.
[8] 顾婷婷, 黄亦露, 王亚男, 等. 基于多头注意力机制的ResNet-UNet短期风电功率预测[J]. 太阳能学报, 2026, 47(3): 474-480.
Gu T T, Huang Y L, Wang Y N, et al.ResNet-UNet short-term wind power prediction with incorporation of multi-head attention mechanism[J]. Acta Energiae Solaris Sinica, 2026, 47(3): 474-480.
[9] 朱童, 王彦沣, 叶希, 等. 考虑多风电机组关联特性的超短期功率预测方法[J]. 四川电力技术, 2025, 48(1): 23-31, 71.
Zhu T, Wang Y F, Ye X, et al.Ultra-short-term power prediction method considering correlation characteristics of multiple wind turbines[J]. Sichuan Electric Power Technology, 2025, 48(1): 23-31, 71.
[10] 张开伟, 文中, 杨生鹏, 等. 考虑风电机组健康状况与双重注意力机制CNN-BiLSTM的超短期功率预测[J]. 可再生能源, 2025, 43(2): 217-224.
Zhang K W, Wen Z, Yang S P, et al.Considering the health status of wind turbines and the dual attention mechanism CNN-BiLSTM ultra-short-term power prediction[J]. Renewable Energy Resources, 2025, 43(2): 217-224.
[11] Wang S X, Shi J R, Yang W, et al.High and low frequency wind power prediction based on Transformer and BiGRU-Attention[J]. Energy, 2024, 288: 129753.
[12] Zhang X, Ye J, Gao L T, et al.Short-term wind power prediction based on ICEEMDAN decomposition and BiTCN-BiGRU-multi-head self-attention model[J]. Electrical Engineering, 2025, 107(3): 2645-2662.
[13] Zhang Y J, Zhang L, Sun D, et al.Short-term wind power forecasting based on VMD and a hybrid SSA-TCN-BiGRU network[J]. Applied Sciences, 2023, 13(17): 9888.
[14] Chen W, Huang H Q, Ma X K, et al.The short-term wind power prediction based on a multi-layer stacked model of BOCNN-BiGRU-SA[J]. Digital Signal Processing, 2025, 156: 104838.
[15] Chi D W, Yang C Z.Wind power prediction based on WT-BiGRU-attention-TCN model[J]. Frontiers in Energy Research, 2023, 11: 1156007.
[16] Meng Y Y, Chang C, Huo J Y, et al.Research on ultra-short-term prediction model of wind power based on attention mechanism and CNN-BiGRU combined[J]. Frontiers in Energy Research, 2022, 10: 920835.
[17] Lundberg S M, Lee S I.A unified approach to interpreting model predictions[C]//Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook, NY, USA, 2017: 4768-4777.
[18] NB/T 31046—2022 风电功率预测系统功能规范[S].
NB/T 31046—2022: Function specification of wind power forecasting system[S].
基金
国家电力投资集团有限公司-上海交通大学“未来能源计划联合基金”（WH410245001/004）
PDF(6540 KB)
