结合时空局部注意力网络与LSTM的日前光伏功率预测模型

宿连超; 樊静雨; 只德天; 吴凯

doi:10.13296/j.1001-1609.hva.2026.08.022

您当前的位置：

首页 >

文章列表页 >

结合时空局部注意力网络与LSTM的日前光伏功率预测模型

研究与分析 | 更新时间：2026-08-12

- 结合时空局部注意力网络与LSTM的日前光伏功率预测模型
- Day-ahead Photovoltaic Power Forecasting Model Based on Spatio-temporal Local Attention Network and LSTM
- 高压电器 2026年62卷第8期页码：180-190
- 作者机构：
  
  国网潍坊供电公司，山东潍坊 261000
  国网山东省电力公司，济南 250013
  天津国投津能发电有限公司，天津 300450
  华北电力大学新能源电力系统国家重点实验室，北京 102206
- 作者简介：
  
  宿连超（1986—），男，高级工程师，研究方向为基于大数据和人工智能的电力系统应用（通信作者）（E-mail：wssg878@sina.com）。
- 基金信息：
  
  国家自然科学基金资助项目;National Natural Science Foundation of China(52007174)
- DOI：10.13296/j.1001-1609.hva.2026.08.022
  中图分类号：
- 收稿：2026-02-22，
  
  修回：2026-04-26，
  
  纸质出版：2026-08-16
- 稿件说明：
移动端阅览
宿连超, 樊静雨, 只德天, 等. 结合时空局部注意力网络与LSTM的日前光伏功率预测模型[J]. 高压电器, 2026,62(8):180-190.

SU Lianchao, FAN Jingyu, ZHI Detian, et al. Day-ahead Photovoltaic Power Forecasting Model Based on Spatio-temporal Local Attention Network and LSTM[J]. High Voltage Apparatus, 2026, 62(8): 180-190.
宿连超, 樊静雨, 只德天, 等. 结合时空局部注意力网络与LSTM的日前光伏功率预测模型[J]. 高压电器, 2026,62(8):180-190. DOI： 10.13296/j.1001-1609.hva.2026.08.022.

SU Lianchao, FAN Jingyu, ZHI Detian, et al. Day-ahead Photovoltaic Power Forecasting Model Based on Spatio-temporal Local Attention Network and LSTM[J]. High Voltage Apparatus, 2026, 62(8): 180-190. DOI： 10.13296/j.1001-1609.hva.2026.08.022.

摘要

光伏功率随时间的波动具有不确定性和随机性，准确可靠的光伏功率预测对电网的实时调度和频率调节具有重要意义。为了提高光伏功率预测的准确性，提出一种结合时空局部注意力网络（spatio-temporal local attention network，STLAN）和长短时记忆网络（long short term memory，LSTM）的日前光伏发电功率预测模型（STLAN-LSTM）。所提STLAN-LSTM由双自注意力机制组成，通过时序自注意力机制（temporal selfattention，TSA）动态定位相似的历史光伏功率序列，同时采用空间自注意力机制（spatial self-attention，SSA）自动捕获相邻站点匹配的光伏功率序列。然后，通过改进的相似序列分数估计策略统计两个注意力通道的输出，使得模型侧重于日常特征中相似局部序列。最后，使用极限学习机（extreme learning machine，ELM）将注意力通道的输出转换为非线性空间，并输入至LSTM网络实现功率预测。在光伏电站测量数据上的实验结果表明，STLAN-LSTM的预测指标MAE、RMSE、MAPE值分别为1.26%、1.61%、23.21%，具有较高的预测准确性，为日前光伏发电功率预测提供有效的技术支持。

Abstract

The fluctuation of photovoltaic(PV)power over time is characterized by uncertainty and randomness. Accurate and reliable PV power forecasting is of great significance for real-time grid dispatch and frequency regulation. For improving the accuracy of PV power forecasting

a day-ahead PV power forecasting model STLAN-LSTM) based on Spatiotemporal Local Attention Network and Long Short-Term Memory(STLAN-LSTM)is proposed. The proposedSTLAN-LSTMis composed of a dual self-attention mechanism. First

the similar historical PV power sequences is dynamically located through the Temporal Self-Attention(TSA)and

at the same time

the spatial self-attention(SSA)mechanism is used to automatically capture matching photovolaic power sequence from adjacent stations. Then

an improved similarity sequenc score estimation strategy is used to aggregate the output of the two attention channels

enabling the model to fous on similar local sequence within the daily features. Finally

the outputs of the attention channels are transformed into a nonlinear space using Extreme Learning Machine(ELM)

which are then fed into an LSTM network to achieve power forecasting . Experimental results on measured data from PV power plants show that the the forecasting indicatorsMAE

RMSE

and MAPE of STLAN-SLTM are 1.26%

1.61%

and 23.21% respectively

indicating high forecasting accuracy and providing effective technical support and day-ahead PV power forecasting.

关键词

Keywords

references

Yao Tiechui, Wang Jue, Wu Haoyan, et al. Intra-hour photovoltaic generation forecasting based on multi-source data and deep learning methods[J]. IEEE Transactions on Sustainable Energy,2022,13 (1):607-618.

徐其春,李良杰,卢泽汉.电网接入光伏发电功率的实时调度与预测研究[J].电网与清洁能源,2023,39(12):141-147.

Xu Qichun, Li Liangjie, Lu Zehan. A study on the real-time scheduling and prediction of photovoltaic power connected to the power grid[J]. Advances of Power System & Hydroelectric Engineering,2023,39(12):141-147.

冯俊琨,王黎光.基于马尔可夫链的智能微电网光伏发电预测[J].计算机应用与软件,2023,40(4):343-349.

Feng Junkun, Wang Liguang. Photovoltaic power generation prediction of smart microgrid based on markov chain[J]. Computer Applications and Software,2023,40(4):343-349.

仇晨光,倪杰,王博仑,等.基于相似日与数据机理混合驱动的光伏发电功率预测[J]. 电网与清洁能源,2025,41(9):101-109.

Qiu Chenguang, Ni Jie, Wang Bolun, et al. Photovoltaic power generation forecasting based on hybrid-driven approach combining similar days and data mechanism[J]. Power System and Clean Energy,2025,41(9):101-109.

刘庆,王有军,张垚,等.基于注意力机制的融合式NCP-DCNN短期光伏功率预测方法[J]. 智慧电力,2025,53(8):62-69.

Liu Qing, Wang Youjun, Zhang Yao, et al. Attention mechanism-based integratedNCP-DCNNmethod for short-term photovoltaic power forecasting[J]. Smart Power,2025,53(8):62-69.

Antonanzas J, Osorio N, Escobar R, et al. Review of photovoltaic power forecasting[J]. Solar Energy,2016(136):78-111.

吐松江·卡日,吴现,马小晶,等.基于地基云图数据多维特征融合的光伏功率预测算法[J]. 电力系统保护与控制,2025,53 (11):84-94.

Tusongjiang Kari, Wu Xian, Ma Xiaojing, et al. Photovoltaic power prediction algorithm based on multidimensional features fusion of ground - based cloud images[J]. Power System Protection and Control,2025,53(11):84-94.

Kim H , Lee D. Probabilistic solar power forecasting based on bivariate conditional solar irradiation distributions[J]. IEEE Transactions on Sustainable Energy,2021,12(4):2031-2041.

王健,刘汇塬,张占喜,等.基于时空特征提取与跨模态融合的光伏集群功率预测[J]. 电力建设,2025,46(11):121-129.

Wang Jian, Liu Huiyuan, Zhang Zhanxi, et al. Power prediction of photovoltaic clusters based on spatio-temporal feature extraction and cross-modal fusion[J]. Electric Power Construction,2025,46 (11):121-129.

史文瑜,张珍翼,杨德昌.基于时间序列大模型TimeGPT光伏功率预测方法研究[J]. 电力科学与技术学报,2025,40(4):150-160.

Shi Wenyu, Zhang Zhenyi, Yang Dechang. Research on photovoltaic power forecasting method based on time series large model TimeGPT[J]. Journal of Electric Power Science and Technology,2025,40(4):150-160.

Chen Changsong, Duan Shanxu, Cai Tao, et al. Online 24 h solar power forecasting based on weather type classification using artificial neural network[J]. Solar Energy,2011,85(11):2856-2870.

Li Hui, Ren Zhouyang, Xu Yan, et al. A multi-data driven hybrid learning method for weekly photovoltaic power scenario forecast[J]. IEEE Transactions on Sustainable Energy,2022,13(1):91-100.

Feng Cong, Cui Mingjian, Hodge B M, et al. Unsupervised clustering-based short-term solar forecasting[J]. IEEE Transactions on Sustainable Energy,2019,10(4):2174-2185.

Zhou Nan, Xu Xiaoyuan, Yan Zheng, et al. Spatio-temporal probabilistic forecasting of photovoltaic power based on monotone broad learning system and copula theory[J]. IEEE Transactions on Sustainable Energy,2022,13(4):1874-1885.

Zhang Ruiyuan, Ma Hui, Saha T K, et al. Photovoltaic nowcasting with bi-level spatio-temporal analysis incorporating sky images[J]. IEEE Transactions on Sustainable Energy,2021,12(3):1766-1776.

Yang Dazhi, Gueymard C. Probabilistic post-processing of gridded atmospheric variables and its application to site adaptation of shortwave solar radiation[J]. Solar Energy,2021,225(1):427-443.

Yagli G M, Yang Dazhi, Srinivasan D. Ensemble solar forecasting and post-processing using dropout neural network and information from neighboring satellite pixels[J]. Renewable and Sustainable Energy Reviews,2022 (155):111909.

Zhang Ruiyuan, Ma Hui, Hua Wen, et al. Data-driven photovoltaic generation forecasting based on a bayesian network with spatialtemporal correlation analysis[J]. IEEE Transactions on Industrial Informatics,2020,16(3):1635-1644.

Agoua X, Girard R, Kariniotakis G. Probabilistic models for spatiotemporal photovoltaic power forecasting[J]. IEEE Transactions on Sustainable Energy,2019,10(2):780-789.

Bessa R, Trindade A, Miranda V. Spatial-temporal solar power forecasting for smart grids[J]. IEEE Transactions on Industrial Informatics,2015,11(1):232-241.

Qin Jun, Jiang Hou, Lu Ning, et al. Enhancing solar PV output forecast by integrating ground and satellite observations with deep learning[J]. Renewable and Sustainable Energy Reviews,2022 (167):112680.

Simeunović J, Schubnel B, Alet P J, et al. Spatio-temporal graph neural networks for multi-site PV power forecasting[J]. IEEE Transactions on Sustainable Energy,2022,13(2):1210-1220.

王海军,居蓉蓉,董颖华.基于时空关联特征与B-LSTM模型的分布式光伏功率区间预测[J].中国电力,2024,57(7):74-80.

Wang Haijun, Ju Rongrong, Dong Yinghua. Distributed photovoltaic power interval prediction based on spatio-temporal correlation feature and B-LSTM model[J]. Electric Power,2024,57(7):74-80.

宋绍剑,李博涵.基于LSTM网络的光伏发电功率短期预测方法的研究[J].可再生能源,2021,39(5):594-602.

Song Shaojian, Li Bohan. Short-term forecasting method of photovoltaic power based onLSTM[J]. Renewable Energy Resources, 2021,39(5):594-602.

Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[EB/OL].[2026-02-22]. https://arxiv. org/abs/1706. 03762.

Zhou Hangxia, Zhang Yujin, Yang Lingfan, et al. Short-term photovoltaic power forecasting based on long short term memory neural network and attention mechanism[J]. IEEE Access,2019 (7):78063-78074.

Shih S Y, Sun Fankeng, Lee H Y. Temporal pattern attention for multivariate time series forecasting[J]. Machine Learning,2019, 108(8):1421-1441.

Gers F A, Schmidhuber J, Cummins F. Learning to forget: continual prediction with LSTM[J]. Neural Computation,2000,12(10):2451-2471.

王瑞,高强,逯静.基于CEEMDAN-LSSVM-ARIMA模型的短期光伏功率预测[J].传感器与微系统,2022,41(5):118-122.

Wang Rui, Gao Qiang, Lu Jing. Short-term photovoltaic power prediction based on CEEMDAN - LSSVM - ARIMA model[J]. Transducer and Microsystem Technologies,2022,41(5):118-122.

周鑫,李燕,曾永辉,等.基于SARIMAX-SVR的光伏发电功率预测[J].电力系统及其自动化学报,2024,36(5):1-8.

Zhou Xin, Li Yan, Zeng Yonghui, et al. Forecasting of photovoltaic power generation based onSARIMAX-SVR[J]. Proceedings of the CSU-EPSA,2024,36(5):1-8.

Lin Ziwei, Li Mengting, Zheng Zihan, et al. Self - attention ConvLSTM for spatiotemporal prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence,2020:11531-11538.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于注意力—门控循环—时间卷积网络的变压器油温预测

基于改进YOLOv5s的绝缘子缺陷检测方法

改进YOLOv11n在输电线路绝缘子主要缺陷检测中的应用研究

基于改进YOLOv7的多尺度特征提取绝缘子缺陷检测算法

基于GE-YOLO算法的输电线路绝缘子多类型缺陷检测研究