Ultra-short Term PV Power Prediction Based on Dynamic Attention Mechanism and Multi-scale Feature Fusion

MA Dong; HUANG Yiling; ZHANG Zhe; XU Sida

doi:10.19805/j.cnki.jcspe.2025.240535

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Ultra-short Term PV Power Prediction Based on Dynamic Attention Mechanism and Multi-scale Feature Fusion

更新时间：2026-10-10

- Ultra-short Term PV Power Prediction Based on Dynamic Attention Mechanism and Multi-scale Feature Fusion
- Chinese Journal of Power Engineering Vol. 45, Issue 11, Pages: 1855-1863(2025)
- 作者机构：
  
  1. 华电电力科学研究院有限公司,浙江,杭州,310030
  2. 华北电力大学控制与计算机工程学院,北京,102206
- 作者简介：
- 基金信息：
- DOI：10.19805/j.cnki.jcspe.2025.240535
  CLC：
- Online First：17 November 2025，
  
  Published：17 November 2025
- 稿件说明：
移动端阅览
马东,黄仪灵,张哲,徐思达. 基于动态注意力机制和多尺度特征融合的超短期光伏发电功率预测动力工程学报, 2025, 45(11): 1855-1863 https://doi.

org/10.19805/j.cnki.jcspe.2025.240535
马东,黄仪灵,张哲,徐思达. 基于动态注意力机制和多尺度特征融合的超短期光伏发电功率预测动力工程学报, 2025, 45(11): 1855-1863 https://doi. DOI： 10.19805/j.cnki.jcspe.2025.240535.

org/10.19805/j.cnki.jcspe.2025.240535 DOI：

摘要

为提升超短期光伏发电功率预测的精度和鲁棒性

提出一种基于Transformer的预测模型。首先

该模型采用小波分解将原始功率序列分解为多尺度分量;其次

通过Transformer模型子模块提取各分量特征

并利用多头注意力机制进行融合;然后

设计了动态注意力机制

依据外部气象条件动态调整注意力权重;最后

采用均方误差和对数似然损失的混合损失函数进行模型训练。结果表明:该方法在预测精度和泛化能力方面均优于现有主流模型。在数据波动较大的夏季

该方法的平均绝对误差为0.085 9 MW

均方根误差为0.115 2 MW

明显低于性能次优的Informer模型

为光伏发电功率的准确预测提供了新的思路和方法。

Abstract

To improve the accuracy and robustness of ultra-short term photovoltaic (PV) power prediction

a Transformer-based prediction model was proposed. Firstly

the wavelet decomposition was employed to break down the original power sequence into multi-scale components. Secondly

features were extracted from each component through sub-modules of the Transformer model

and above features were fused by multi-head attention mechanism. Then

a dynamic attention mechanism was designed to dynamically adjust attention weights based on external meteorological conditions. Finally

a hybrid loss function combining mean squared error and log-likelihood loss was adopted for model training. Results indicate that the proposed method outperforms existing mainstream models in terms of both prediction accuracy and generalization capability. During summer

when data exhibit significant fluctuations

this method can achieve an average absolute error of 0.085 9 MW and a root mean square error of 0.115 2 MW

which are notably lower than those of the sub-optimal Informer model. It provides new ideas and methods for the accurate prediction of PV power generation.

关键词

Keywords

references

张曼铮, 郭伟, 方宇成, 等. 基于太阳能分频利用的光伏/光热综合发电系统热力学分析[J]. 动力工程学报, 2024, 44(8): 1216-1225. ZHANG Manzheng, GUO Wei, FANG Yucheng, et al. Thermodynamic analysis of a photovoltaic/photothermal integrated power generation system based on solar energy spectrum utilization[J]. Journal of Chinese Society of Power Engineering, 2024, 44(8): 1216-1225.

田亮, 王冠杰. 基于模型-数据联合的光伏-光热系统储能量预测[J]. 动力工程学报, 2024, 44(6): 911-918. TIAN Liang, WANG Guanjie. Energy storage prediction of photovoltaic-concentrating solar power system based on model and data[J]. Journal of Chinese Society of Power Engineering, 2024, 44(6): 911-918.

董存, 王铮, 白捷予, 等. 光伏发电功率超短期预测方法综述[J]. 高电压技术, 2023, 49(7): 2938-2951. DONG Cun, WANG Zheng, BAI Jieyu, et al. Review of ultra-short-term forecasting methods for photovoltaic power generation[J]. High Voltage Engineering, 2023, 49(7): 2938-2951.

周建国, 周路明, 王剑宇, 等. 基于优化分解与误差修正的光伏发电功率预测[J]. 太阳能学报, 2024, 45(7): 502-509. ZHOU Jianguo, ZHOU Luming, WANG Jianyu, et al. PV power forecasting based on optimal decomposition and error correction[J]. Acta Energiae Solaris Sinica, 2024, 45(7): 502-509.

陈庆明, 廖鸿飞, 孙颖楷, 等. 基于GWO-GRU的光伏发电功率预测[J]. 太阳能学报, 2024, 45(7): 438-444. CHEN Qingming, LIAO Hongfei, SUN Yingkai, et al. Photovoltaic power prediction model based on GWO-GRU[J]. Acta Energiae Solaris Sinica, 2024, 45(7): 438-444.

黄牧涛, 邢芳菲, 陈兴邦, 等. 基于K-means聚类和极限学习机组合算法的短期光伏功率预测[J]. 水电能源科学, 2024, 42(2): 217-220, 216. HUANG Mutao, XING Fangfei, CHEN Xingbang, et al. Short-term PV power prediction based on K-means clustering and extreme learning machine combination algorithm[J]. Science of Hydropower and Energy, 2024, 42(2): 217-220, 216.

李立刚, 孙立群, 戴永寿, 等. 基于时空图卷积神经网络的光伏发电功率超短期预测方法[J]. 热能动力工程, 2023, 38(9): 152-157, 173. LI Ligang, SUN Liqun, DAI Yongshou, et al. Ultra-short-term prediction method of photovoltaic power based on spatio-temporal graph convolutional neural network[J]. Journal of Engineering for Thermal Energy and Power, 2023, 38(9): 152-157, 173.

刘国海, 孙文卿, 吴振飞, 等. 基于Attention-GRU的短期光伏发电功率预测[J]. 太阳能学报, 2022, 43(2): 226-232. LIU Guohai, SUN Wenqing, WU Zhenfei, et al. Short-term photovoltaic power forecasting based on Attention-GRU model[J]. Acta Energiae Solaris Sinica, 2022, 43(2): 226-232.

LPEZ SANTOS M, GARCA-SANTIAGO X, ECHEVARRA CAMARERO F, et al. Application of temporal fusion transformer for day-ahead PV power forecasting[J]. Energies, 2022, 15(14): 5232.

JU Yun, LI Jing, SUN Guangyu. Ultra-short-term photovoltaic power prediction based on self-attention mechanism and multi-task learning[J]. IEEE Access, 2020, 8: 44821-44829.

LIU Jingxuan, ZANG Haixiang, CHENG Lilin, et al. A transformer-based multimodal-learning framework using sky images for ultra-short-term solar irradiance forecasting[J]. Applied Energy, 2023, 342: 121160.

黄莉, 甘恒玉, 刘兴举, 等. 基于Transformer编码器的超短期光伏发电功率预测[J]. 智慧电力, 2024, 52(5): 16-22, 59. HUANG Li, GAN Hengyu, LIU Xingju, et al. Ultra-short-term photovoltaic power generation prediction based on transformer encoder[J]. Smart Power, 2024, 52(5): 16-22, 59.

张倩, 蒙飞, 李涛, 等. 基于周期信息增强的Informer光伏发电功率预测[J]. 中国电力, 2023, 56(10): 186-193. ZHANG Qian, MENG Fei, LI Tao, et al. Informer photovoltaic power generation forecasting based on cycle information enhancement[J]. Electric Power, 2023, 56(10): 186-193.

刘世鹏, 宁德军, 马崛. 针对光伏发电功率预测的LSTformer模型[J]. 计算机工程与应用, 2024, 60(9): 317-325. LIU Shipeng, NING Dejun, MA Jue. LSTformer model for photovoltaic power prediction[J]. Computer Engineering and Applications, 2024, 60(9): 317-325.

TAO Kejun, ZHAO Jinghao, TAO Ye, et al. Operational day-ahead photovoltaic power forecasting based on transformer variant[J]. Applied Energy, 2024, 373: 123825.

WANG Yuqing, FU Wenjie, WANG Junlong, et al. Ultra-short-term distributed PV power forecasting for virtual power plant considering data-scarce scenarios[J]. Applied Energy, 2024, 373: 123890.

CAO Yisheng, LIU Gang, LUO Donghua, et al. Multi-timescale photovoltaic power forecasting using an improved stacking ensemble algorithm based LSTM-informer model[J]. Energy, 2023, 283: 128669.

董坤, 冉鹏, 刘旭, 等. 基于权参数优化的并行深度学习光伏功率预测[J]. 动力工程学报, 2024, 44(1): 91-98. DONG Kun, RAN Peng, LIU Xu, et al. Photovoltaic power prediction by weight parameter optimization-based parallel deep learning[J]. Journal of Chinese Society of Power Engineering, 2024, 44(1): 91-98.

陆王琳, 陆启亮, 张志洪. 碳中和背景下综合智慧能源发展趋势[J]. 动力工程学报, 2022, 42(1): 10-18. LU Wanglin, LU Qiliang, ZHANG Zhihong. An overview of the integrated energy systems' development under the background of carbon neutralization[J]. Journal of Chinese Society of Power Engineering, 2022, 42(1): 10-18.

张姗, 冬雷, 纪德洋, 等. 基于NWP相似性分析的超短期光伏发电功率预测[J]. 太阳能学报, 2022, 43(4): 142-147. ZHANG Shan, DONG Lei, JI Deyang, et al. Power forecasting of ultra-short-term photovoltaic station based on NWP similarity analysis[J]. Acta Energiae Solaris Sinica, 2022, 43(4): 142-147.

谭海旺, 杨启亮, 邢建春, 等. 基于XGBoost-LSTM组合模型的光伏发电功率预测[J]. 太阳能学报, 2022, 43(8): 75-81. TAN Haiwang, YANG Qiliang, XING Jianchun, et al. Photovoltaic power prediction based on combined XGBoost-LSTM model[J]. Acta Energiae Solaris Sinica, 2022, 43(8): 79-81.

孙子涵, 姜志海, 刘延龙, 等. 基于小波变换和优化的Elman神经网络的光伏功率预测[J]. 电网与清洁能源, 2022, 38(6): 98-103, 112. SUN Zihan, JIANG Zhihai, LIU Yanlong, et al. Photovoltaic power prediction based on wavelet transform and optimized Elman neural network[J]. Power System and Clean Energy, 2022, 38(6): 98-103, 112.

LI Yifan, LIU Gang, CAO Yisheng, et al. WNPS-LSTM-Informer: a hybrid stacking model for medium-term photovoltaic power forecasting with ranked feature selection[J]. Renewable Energy, 2025, 244: 122687.

LI Ning, DONG Jie, LIU Lingyue, et al. A novel EMD and causal convolutional network integrated with transformer for ultra short-term wind power forecasting[J]. International Journal of Electrical Power & Energy Systems, 2023, 154: 109470.

XU Shijie, ZHANG Ruiyuan, MA Hui, et al. On vision transformer for ultra-short-term forecasting of photovoltaic generation using sky images[J]. Solar Energy, 2024, 267: 112203.

吴艳娟, 荣旺, 郭玥, 等. 基于SVMD-IDBO-KELM的短期光伏发电功率预测[J]. 太阳能学报, 2025, 46(6): 271-279. WU Yanjuan, RONG Wang, GUO Yue, et al. Short-term photovoltaic power forecasting based on SVMD-IDBO-KELM[J]. Acta Energiae Solaris Sinica, 2025, 46(6): 271-279.

李宏扬, 高丙朋. 基于改进VMD和SNS-Attention-GRU的短期光伏发电功率预测[J]. 太阳能学报, 2023, 44(8): 292-300. LI Hongyang, GAO Bingpeng. Short-term PV power forecasting based on improved VMD and SNS-Attention-GRU[J]. Acta Energiae Solaris Sinica, 2023, 44(8): 292-300.

Views

582

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Dynamic Modeling of SNCR Denitrification System for Waste Incinerators Based on BKA-Transformer-BiGRU

Research on the Thermoelectric Performance of a PVT System Based on a Full-flow Channel Cooling Collector

Self-triggered Pitch Control for Wind Turbines Based on Semi-Markov

Research on the Application of Singular Value Decomposition in Early Friction Fault Diagnosis

Effect of Fractal Characteristics on Dynamic Response of Bionic Floating Wind Turbine Platform

Related Author

No data

Related Institution

No data

AI问答

⁰