Research on Photovoltaic Power Prediction Based on Multimodal Fusion of Ground Cloud Map and Meteorological Factors

DENG Fangming; LIU Tao; WANG Jinbo; GAO Bo; WEI Baoquan; LI Zewen

doi:10.13334/j.0258-8013.pcsee.240941

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Research on Photovoltaic Power Prediction Based on Multimodal Fusion of Ground Cloud Map and Meteorological Factors

更新时间：2026-07-27

- Research on Photovoltaic Power Prediction Based on Multimodal Fusion of Ground Cloud Map and Meteorological Factors
- Vol. 45, Issue 18, Pages: 7193-7205(2025)
- 作者机构：
  
  华东交通大学电气与自动化工程学院,江西省,南昌市,330013
- 作者简介：
- 基金信息：
- DOI：10.13334/j.0258-8013.pcsee.240941
  CLC：
- Published：2025
- 稿件说明：
移动端阅览
DENG Fangming, LIU Tao, WANG Jinbo, et al. Research on Photovoltaic Power Prediction Based on Multimodal Fusion of Ground Cloud Map and Meteorological Factors[J]. 2025, 45(18): 7193-7205.
DOI：

DENG Fangming, LIU Tao, WANG Jinbo, et al. Research on Photovoltaic Power Prediction Based on Multimodal Fusion of Ground Cloud Map and Meteorological Factors[J]. 2025, 45(18): 7193-7205. DOI： 10.13334/j.0258-8013.pcsee.240941.

摘要

现有地基云图与气象因素融合方法难以充分利用地基云图和气象因素两种模态之间的相关性和互补性，导致无法有效提升辐照度变化剧烈天气下的光伏功率预测准确度。该文提出一种基于地基云图与气象因素多模态融合的光伏功率预测方法。采用一种改进Transformer网络，通过将气象数据输入部分的位置编码替换为时序卷积网络(time series convolutional network，TCN)，提升网络对气象数据的特征提取能力；通过将网络解码器部分的多头注意力机制模块替换为长短时记忆网络(long short-term memory networks，LSTM)，提升网络对时序序列的预测能力。分别引入引导注意力机制和低秩多模态融合算法对云图特征和气象特征进一步特征提取和融合，以充分利用不同源数据之间的相关性和互补性。结果表明，上述方法的均方根误差(root mean square error，RMSE)、平均绝对误差(mean absolute error，MAE)和拟合优度(R2)分别为0.294、0.248和0.866，可以提升辐照度变化剧烈天气下的光伏功率预测准确度，从而提高电力系统运行的稳定性和可靠性。

Abstract

The existing fusion methods of foundation cloud maps and meteorological factors are difficult to fully explore the correlation and complementarity between the two modes

resulting in the inability to effectively improve the accuracy of photovoltaic power prediction under severe irradiance changes in weather. In response to the above issues

this article proposes a photovoltaic power prediction method based on multimodal fusion of ground cloud maps and meteorological factors. Adopting an improved Transformer network by replacing the position encoding of the meteorological data input with a time series convolutional network(TCN) to enhance the network's feature extraction capability for meteorological data; By replacing the multi head attention mechanism module in the network decoder with a long short-term memory networks(LSTM) network module

the network's ability to predict temporal sequences is improved. By introducing guided attention mechanism and low rank multimodal fusion algorithm separately

further feature extraction and fusion of cloud map features and meteorological features can fully utilize the correlation and complementarity between different source data. The simulation results show that the average root mean square error (RMSE)

mean absolute error (MAE)

and R2 values of the above method are 0.294

0.248

and 0.866 respectively

which improving the accuracy of photovoltaic power prediction under severe irradiance changes in weather.

关键词

Keywords

references

Views

1171

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Photovoltaic Ultra-short-term Multi-step Power Prediction Based on Satellite Cloud Image and Station Spatial Distribution

计及复杂气象耦合特性的模块化去噪变分自编码器多源–荷联合场景生成

基于云图特征提取的改进混合神经网络超短期光伏功率预测方法

基于气象资源插值与迁移学习的广域分布式光伏功率预测方法

Real-time Optimal Scheduling of Cascaded Hydropower Stations Under High-intensity Peak Shaving and Frequency Regulation

Related Author

YANG Mao

ZHAO Yuting

WANG Maochun

ZHANG Jietan

ZHU Han

CUI Yang

黄南天

Related Institution

Key Laboratory of Modern Power System Simulation and Control & Renewable Energy Technology, Ministry of Education (Northeast Electric Power University)

现代电力系统仿真控制与绿色电能新技术教育部重点实验室(东北电力大学)

上海电力大学电气工程学院

国网陕西省电力有限公司电力调度控制中心

AI问答

⁰