基于相似日聚类和时序图像特征提取的光伏功率预测方法

郭威; 徐立; 汤旭晶; 赵丹阳; 陈昉林; 汪恬

doi:10.19912/j.0254-0096.tynxb.2024-2355

PDF(3009 KB)

太阳能学报 ›› 2026, Vol. 47 ›› Issue (5) : 650-659. DOI: 10.19912/j.0254-0096.tynxb.2024-2355

基于相似日聚类和时序图像特征提取的光伏功率预测方法

郭威¹, 徐立¹, 汤旭晶^1~3, 赵丹阳¹, 陈昉林¹, 汪恬¹

作者信息 +

PHOTOVOLTAIC POWER PREDICTION METHOD BASED ON SIMILAR DAY CLUSTERING AND TEMPORAL IMAGE FEATURE EXTRACTION

Guo Wei¹, Xu Li¹, Tang Xujing^1~3, Zhao Danyang¹, Chen Fanglin¹, Wang Tian¹

Author information +

文章历史 +

摘要

提出一种融合时序图像特征提取的混合深度学习光伏功率预测模型。首先,基于综合相关系数法构建多维气象特征筛选机制,实现关键影响因子的优化选择。其次,提出融合LB_Keogh距离的动态时间规整（DTW）度量方法,对K-中心点聚类算法进行改进,增强了算法的鲁棒性和聚类效率,实现了晴天、多云和阴雨天气模式的准确划分。最后,将选定的高相关性气象特征变量和历史光伏功率数据转换为二维格拉姆角场图像,输入到CSWin-Transformer模型进行功率预测。与对比模型相比,所提模型在3种典型天气工况下均表现出更高的预测精度。

Abstract

This paper introduces a hybrid deep learning PV power prediction model that leverages feature extraction from time-series images. Firstly, a multi-dimensional meteorological feature screening mechanism was developed using the comprehensive correlation coefficient method to optimize the selection of key influencing factors. Secondly, an enhanced K-Medoids clustering algorithm was proposed by incorporating a dynamic time warping （DTW） distance measure based on LB_Keogh distance, which improves the robustness and clustering efficiency of the algorithm and enables precise classification of sunny, cloudy, and rainy weather patterns. Finally, the selected high-correlation meteorological variables and historical PV power data were transformed into two-dimensional Gramian angular field images and fed into the CSWin-Transformer model for power prediction. Compared with other models, the proposed approach demonstrates superior prediction accuracy under three typical weather conditions, offering a novel methodology for PV power generation prediction.

导出引用

郭威, 徐立, 汤旭晶, 赵丹阳, 陈昉林, 汪恬. 基于相似日聚类和时序图像特征提取的光伏功率预测方法[J]. 太阳能学报. 2026, 47(5): 650-659 https://doi.org/10.19912/j.0254-0096.tynxb.2024-2355

Guo Wei, Xu Li, Tang Xujing, Zhao Danyang, Chen Fanglin, Wang Tian. PHOTOVOLTAIC POWER PREDICTION METHOD BASED ON SIMILAR DAY CLUSTERING AND TEMPORAL IMAGE FEATURE EXTRACTION[J]. Acta Energiae Solaris Sinica. 2026, 47(5): 650-659 https://doi.org/10.19912/j.0254-0096.tynxb.2024-2355

中图分类号： TM615

参考文献

[1] STRAUB N, HERZBERG W, DITTMANN A, et al.Blending of a novel all sky imager model with persistence and a satellite based model for high-resolution irradiance nowcasting[J]. Solar energy, 2024, 269: 112319.
[2] 王迎春, 王志硕, 刘洋, 等. 基于联邦学习的海上分布式光伏超短期功率预测[J]. 控制与决策, 2025, 40(2): 441-450.
WANG Y C, WANG Z S, LIU Y, et al.Ultra-short-term power prediction of offshore distributed PV based on federated learning[J]. Control and decision, 2025, 40(2): 441-450.
[3] 王超, 蔺红, 庞晓虹. 基于HPO-VMD和MISMA-DHKELM的短期光伏功率组合预测[J]. 太阳能学报, 2023, 44(12): 65-73.
WANG C, LIN H, PANG X H.Short-term PV power combination prediction based on HPO-VMD and MISMA-DHKELM[J]. Acta energiae solaris sinica, 2023, 44(12): 65-73.
[4] 刘源延, 孔小兵, 马乐乐, 等. 基于小波包变换与深度学习的超短期光伏功率预测[J]. 太阳能学报, 2024, 45(5): 537-546.
LIU Y Y, KONG X B, MA L L, et al.Ultra-short-term PV power prediction based on wavelet packet transform and deep learning[J]. Acta energiae solaris sinica, 2024, 45(5): 537-546.
[5] 龙小慧, 秦际赟, 张青雷, 等. 基于相似日聚类及模态分解的短期光伏发电功率组合预测研究[J]. 电网技术, 2024, 48(7): 2948-2957.
LONG X H, QIN J Y, ZHANG Q L, et al.Short-term photovoltaic power prediction study based on similar day clustering and modal decomposition[J]. Power system technology, 2024, 48(7): 2948-2957.
[6] 黄泽, 毕贵红, 谢旭, 等. 基于MBI-PBI-ResNet的超短期光伏功率预测[J]. 电力系统保护与控制, 2024, 52(2): 165-176.
HUAN Z, BI G H, XIE X, et al.Ultra-short-term PV power prediction based on MBI-PBI-ResNet[J]. Power system protection and control, 2024, 52(2): 165-176.
[7] WU X, ZHEN Z, ZHANG J, et al.Multidimensional feature extraction based minutely solar irradiance forecasting method using all-sky images[J]. IEEE transactions on industry applications, 2024, 60(3): 4494-4504.
[8] ZHANG X, GAO R, ZHU C, et al.Ultra-short-term prediction of regional photovoltaic power based on dynamic graph convolutional neural network[J]. Electric power systems research, 2024, 226: 109965.
[9] 路志英, 周庆霞, 李鑫, 等. 基于地基云图图像特征的光伏功率预测[J]. 电力系统及其自动化学报, 2020, 32(8): 70-76.
LU Z Y, ZHOU Q X, LL X, et al.Photovoltaic power prediction based on image features of ground-based cloud map[J]. Journal of electric power system and automation, 2020, 32(8): 70-76.
[10] 司志远, 杨明, 于一潇, 等. 基于卫星云图特征区域定位的超短期光伏功率预测方法[J]. 高电压技术, 2021, 47(4): 1214-1223.
SI Z Y, YANG M, YU Y X, et al.Ultra-short-term photovoltaic power prediction method based on satellite cloud image feature area positioning[J]. High voltage engineering, 2021, 47(4): 1214-1223.
[11] 杨国华, 张鸿皓, 郑豪丰, 等. 基于相似日聚类和IHGWO-WNN-AdaBoost模型的短期光伏功率预测[J]. 高电压技术, 2021, 47(4): 1185-1194.
YANG G H, ZHANG H H, ZHENG H F, et al.Short-term PV power prediction based on similar daily clustering and IHGWO-WNN-AdaBoost model[J]. High voltage engineering, 2021, 47(4): 1185-1194.
[12] 万俊良, 罗小燕, 邓涛. 基于LS-DTW和优化k-medoids的磨音信号聚类分析[J]. 噪声与振动控制, 2023, 43(6): 109-116.
WAN J L, LUO X Y, DENG T.Clustering analysis of grinding sound signals based on LS-DTW and optimal k-medoids[J]. Noise and vibration control, 2023, 43(6): 109-116.
[13] 刘金金. 优化初始类中心的自适应K-medoids算法[J]. 河南师范大学学报(自然科学版), 2025(1): 106-115.
LIU J J.Adaptive K-medoids algorithm for optimizing initial class center[J]. Journal of Henan Normal University (natural science edition), 2025(1): 106-115.
[14] CHEN Y, LIU X, LI X, et al.Delineating urban functional areas with building-level social media data: a dynamic time warping (DTW) distance based k-medoids method[J]. Landscape and urban planning, 2017, 160: 48-60.
[15] KEOGH E, RATANAMAHATANA C A.Exact indexing of dynamic time warping[J]. Knowledge and information systems, 2005, 7(3): 358-386.
[16] 李莹, 涂志德, 刘艳芳, 等. 基于时间序列相似度的城市功能区识别研究[J]. 地理空间信息, 2021, 19(1): 22-29,47,4.
LI Y, TU Z D, LIU Y F, et al. Research on urban functional zone identification based on time series similarity[J]. Geospatial information, 2021, 19(1): 22-29,47,4.
[17] 吕伟杰, 方一帆, 程泽. 基于模糊C均值聚类和样本加权卷积神经网络的日前光伏出力预测研究[J]. 电网技术, 2022, 46(1): 231-238.
LYU W J, FANG Y C, CHENG Z.Research on front-ahead photovoltaic output prediction based on fuzzy C-means clustering and sample-weighted convolutional neural network[J]. Power system technology, 2022, 46(1): 231-238.
[18] TIAN W, WU J, CUI H, et al.Drought prediction based on feature-based transfer learning and time series imaging[J]. IEEE access, 2021, 9: 101454-101468.
[19] DONG X, BAO J, CHEN D, et al.CSWin Transformer: a general vision transformer backbone with cross-shaped windows[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA, IEEE, 2022.
[20] DKA Solar Centre[DB/OL]. http://dkasolarcentre.com.au/historical-data/download.