基于双注意力和神经网络的光伏功率预测

摘要
图/表
参考文献
相关文章 (10)

全文: PDF (977 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要针对光伏发电波动性大,时序过长时易丢失重要信息等问题,本文集合注意力机制、卷积神经网络（CNN）和双向长短期记忆（BiLSTM）网络的优势,提出融合双重注意力（DA）机制和神经网络的光伏发电功率预测方法DA-CNN-BiLSTM。首先,引入特征和时间双注意力机制,自主挖掘光伏输出功率与各气象特征和历史关键信息之间的关联关系,赋予各气象特征量不同的权重;其次,利用CNN-BiLSTM混合网络预测;最后,基于实际光伏发电站进行预测实验,结果表明所提方法能较好地反映数据的动态变化,在更长的预测时间尺度和多变的天气下,预测性能提升明显。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	林仁雄
	郭凯琪
	蒋涵
	张锦泉

关键词 ：注意力机制, 双向长短期记忆（BiLSTM）, 光伏发电功率预测

Abstract：In view of the large volatility of photovoltaics and the easy loss of important information when the timing is too long, this paper integrates the advantages of attention mechanism, convolutional neural network (CNN), and bidirectional long short term memory network (BiLSTM) to propose a photovoltaic power prediction method that combines dual attention (DA) mechanism and neural network, which is called dual attention mechanism based CNN-BiLSTM (DA-CNN-BiLSTM). First, the feature and time dual attention mechanism is introduced to independently mine the correlation between photovoltaic power and various meteorological features and historical key information, and different weights are given to each meteorological feature. Then, the prediction is conducted with CNN-BiLSTM. Finally, a prediction experiment based on a real photovoltaic power station is carried out, and results show that the prposed method can better reflect the dynamic changes of the data, and the prediction performance is significantly improved under longer prediction time scales and changeable weather conditions.

Key words： attention mechanism bidirectional long short term memory (BiLSTM) predition of photovoltaic power generation

收稿日期: 2025-11-20

作者简介: 林仁雄（1983—）,男,福建省莆田市人,本科,工程师,主要从事新能源发电预测、电力负荷预测管理工作。

引用本文:

林仁雄, 郭凯琪, 蒋涵, 张锦泉. 基于双注意力和神经网络的光伏功率预测[J]. 电气技术, 2026, 27(6): 17-22. LIN Renxiong, GUO Kaiqi, JIANG Han, ZHANG Jinquan. Photovoltaic power prediction based on dual attention and neural networks. Electrical Engineering, 2026, 27(6): 17-22.

链接本文:

https://dqjs.cesmedia.cn/CN/Y2026/V27/I6/17

[1] 严璐晗, 林培杰, 程树英, 等. 基于增量学习的CNN- LSTM光伏功率预测[J]. 电气技术, 2024, 25(5): 31-40.
[2] 彭程, 徐建勇, 赵书琪, 等. 考虑分层碳排放的多时间尺度光储充一体站日前运行策略[J]. 电气技术, 2025, 26(1): 1-13, 63.
[3] Ahmed R, Sreeram V, Togneri R, et al.Com- putationally expedient photovoltaic power forecasting: a LSTM ensemble method augmented with adaptive weighting and data segmentation technique[J]. Energy Conversion and Management, 2022, 258: 115563.
[4] 刘伟, 杨凯宁. 基于数值天气预报因子扩充和改进集成学习的高寒地区短期光伏功率预测[J]. 电气技术, 2024, 25(8): 1-10, 17.
[5] 武明义, 焦超凡, 瞿博阳, 等. 基于相似日算法及集成学习的短期光伏预测模型[J]. 电气技术, 2021, 22(4): 33-37.
[6] 王志远, 王守相, 陈海文, 等. 考虑空间相关性采用LSTM神经网络的光伏出力短期预测方法[J]. 电力系统及其自动化学报, 2020, 32(5): 78-85.
[7] Wang Kejun, Qi Xiaoxia, Liu Hongda.Photovoltaic power forecasting based LSTM-convolutional net- work[J]. Energy, 2019, 189: 116225.
[8] 谢小瑜, 周俊煌, 张勇军, 等. 基于W-BiLSTM的可再生能源超短期发电功率预测方法[J]. 电力系统自动化, 2021, 45(8): 175-184.
[9] 梁宏涛, 王莹, 刘红菊, 等. 基于注意力机制的CNN- BiGRU短期光伏发电功率预测[J]. 计算机测量与控制, 2022, 30(6): 259-265.
[10] DKA Solar Centre. DKASC, Alice Springs[EB/OL].[2026-02-12]. http://dkasolarcentre.com.au/locations/alice-springs.
[11] 彭昱, 符琛, 郭昕, 等. 基于FCM-BOA-TCN-GRU的分布式光伏出力异常检测方法[J]. 电工技术学报, 2025, 40(17): 5389-5401.
[12] 刘丽桑, 郭凯琪, 陈健, 等. 基于VMD-MWOA-ELM的日前光伏功率预测[J]. 福建工程学院学报, 2023, 21(3): 269-276.
[13] 汤德清, 朱武, 侯林超. 基于CNN-LSTM-XGBoost模型的超短期光伏功率预测[J]. 电源技术, 2022, 46(9): 1048-1052.
[14] 许可, 王珏, 姚铁锤, 等. 基于分时LSTM的光伏发电功率预测[J]. 数据与计算发展前沿, 2021, 3(5): 109-117.
[15] 杨晶显, 张帅, 刘继春, 等. 基于VMD和双重注意力机制LSTM的短期光伏功率预测[J]. 电力系统自动化, 2021, 45(3): 174-182.