针对光伏功率预测存在模型易受干扰、稳定性较差的问题,提出一种基于卷积神经网络（CNN）、双向长短期记忆网络（BiLSTM）和鲁棒注意力机制（RobustAttention）的光伏功率预测方法。利用自适应噪声完备集合经验模态分解（CEEMDAN）将历史功率数据分解为若干固有模态函数（IMF）,从而有效提升数据的稳定性;其次,将IMF与其他特征序列输入到CNN-BiLSTM模块,用以挖掘数据时空特征;最后,将提取到的中间向量输入鲁棒注意力模块,以进一步提高预测精度。经过不同模型实验对比分析,验证该模型具有更高预测精度,相比于使用传统Attention模块,该模型在多个预测时段上的均方根误差（RMSE）指标平均降低8.27%。
Abstract
A photovoltaic power prediction method based on CNN-BiLSTM and an improved robust attention mechanism （RobustAttention） is proposed to address the issues of model susceptibility to disturbances and poor stability. First,adaptive noise complete ensemble empirical mode decomposition with adaptive noise （CEEMDAN） is used to decompose historical power data into several intrinsic mode functions （IMF）,providing more stable data. Next,the IMFs and original data are input into the CNN-BiLSTM-RobustAttention model,where the local feature extraction ability of CNN and the long-term dependency correlation capturing ability of BiLSTM are combined to extract spatiotemporal features from the data. Finally,the features are input into the robust attention module to obtain the prediction results. Ablation experiments using historical power generation data from a photovoltaic power station in Jiangsu are conducted to validate the model’s prediction performance. The traditional attention mechanism is replaced with Robust Attention for comparison. The results show that the proposed method outperforms all models in terms of prediction performance across different time periods,with each component contributing to the improvement of model performance. RobustAttention can better capture anomalies in photovoltaic power generation data,further enhancing model stability and significantly improving prediction accuracy.
关键词
光伏功率预测 /
神经网络 /
深度学习 /
双向长短期记忆网络 /
注意力机制 /
卷积神经网络
Key words
photovoltaic power prediction /
neural network /
deep learning /
bidirectional long short-term memory /
attention mechanism /
convolutional neural network
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 王德文, 焦天媛. 基于二次分解的不同太阳辐射下光伏功率预测[J]. 太阳能学报, 2024, 45(9): 360-368.
Wang D W, Jiao T Y.Prediction of photovoltaic power under different solar radiation based on secondary decomposition[J]. Acta Energiae Solaris Sinica, 2024, 45(9): 360-368.
[2] FALOPE T, LAO L Y, HANAK D, et al.Hybrid energy system integration and management for solar energy: a review[J]. Energy Conversion and Management: x, 2024, 21: 100527.
[3] Fu X Q, Zhang C Y, Xu Y, et al.Statistical machine learning for power flow analysis considering the influence of weather factors on photovoltaic power generation[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(3): 5348-5362.
[4] 时培明, 郭轩宇, 杜清灿, 等. 基于TCN-BiLSTM-Attention-ESN的光伏功率预测[J]. 太阳能学报, 2024, 45(9): 304-316.
Shi P M, Guo X Y, Du Q C, et al.Photovoltaic power prediction based on TCN-BILSTM-Attention-ESN[J]. Acta Energiae Solaris Sinica, 2024, 45(9): 304-316.
[5] Mayer M J, GróF G. Extensive comparison of physical models for photovoltaic power forecasting[J]. Applied Energy, 2021, 283: 116239.
[6] Mayer M J.Benefits of physical and machine learning hybridization for photovoltaic power forecasting[J]. Renewable and Sustainable Energy Reviews, 2022, 168: 112772.
[7] Mayer M J, Yang D Z.Pairing ensemble numerical weather prediction with ensemble physical model chain for probabilistic photovoltaic power forecasting[J]. Renewable and Sustainable Energy Reviews, 2023, 175(C): 113171.
[8] Gupta P, Singh R.PV power forecasting based on data-driven models: a review[J]. International Journal of Sustainable Engineering, 2021, 14(6): 1733-1755.
[9] Ullah K, Ahsan M, Hasanat S M, et al.Short-term load forecasting: a comprehensive review and simulation study with CNN-LSTM hybrids approach[J]. IEEE Access, 2024, 12: 111858-111881.
[10] Zhou M R, Wang L, Hu F, et al.ISSA-LSTM: a new data-driven method of heat load forecasting for building air conditioning[J]. Energy and Buildings, 2024, 321: 114698.
[11] Ahmad Khan Z, Hussain T, Baik S W.Dual stream network with attention mechanism for photovoltaic power forecasting[J]. Applied Energy, 2023, 338: 120916.
[12] Liu W C, Mao Z Z.Short-term photovoltaic power forecasting with feature extraction and attention mechanisms[J]. Renewable Energy, 2024, 226: 120437.
[13] Li G Z, Ding C J, Zhao N N, et al.Research on a novel photovoltaic power forecasting model based on parallel long and short-term time series network[J]. Energy, 2024, 293: 130621.
[14] Zhu J B, Li M R, Luo L, et al.Short-term PV power forecast methodology based on multi-scale fluctuation characteristics extraction[J]. Renewable Energy, 2023, 208: 141-151.
[15] 陈庆斌, 杨耿煌, 耿丽清, 等. 基于相似日选取和数据重构的短期光伏功率组合预测方法[J]. 中国电力, 2024, 57(12): 71-81.
Chen Q B, Yang G H, Geng L Q, et al.Short term photovoltaic power combination prediction method based on similar day selection and data reconstruction[J]. Electric Power, 2024, 57(12): 71-81.
[16] Li Q Y, Wang G S, Wu X R, et al.Arctic short-term wind speed forecasting based on CNN-LSTM model with CEEMDAN[J]. Energy, 2024, 299: 131448.
[17] Gao W Z, Hou Z C, Liu X R, et al.ProTransformer: robustify transformers via plug-and-play paradigm[C]//Advances in Neural Information Processing Systems 37. Vancouver, BC, Canada, 2024: 137557-137609.
[18] Torres M E, Colominas M A, Schlotthauer G, et al.A complete ensemble empirical mode decomposition with adaptive noise[C]//2011 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). Prague, Czech Republic, 2011: 4144-4147.
[19] He Y T, Gao Q Z, Jin Y Y, et al.Short-term photovoltaic power forecasting method based on convolutional neural network[J]. Energy Reports, 2022, 8: 54-62.
[20] 杨轶航, 韩璐, 史华勃, 等. 基于相似日与ISC-BiLSTM的短期光伏功率预测方法[J]. 太阳能学报, 2025, 46(1): 676-685.
Yang Y H, Han L, Shi H B, et al.Short-term photovoltaic power forecast method based on similar days and ISC-BiLSTM[J]. Acta Energiae Solaris Sinica, 2025, 46(1): 676-685.
基金
国家自然科学基金（52171185; 52371194）
PDF(1406 KB)
