基于地基云图时空特征增强的超短期光伏功率预测方法

doi:10.16182/j.issn1004731x.joss.25-0856

摘要/Abstract

摘要：

针对地基云图获取存在时间滞后以及现有光伏功率预测模型精度不足的问题，提出了一种基于地基云图时空特征增强的改进LSTM超短期光伏功率预测方法STFE-GCI-ILSTM (spatial-temporal feature enhancement-ground-based cloud image-improved LSTM)。构建了多尺度时空特征增强的地基云图预测模型(STFE-GCI)，通过空间特征增强、时间特征增强及多尺度特征融合技术，提取历史云图序列的深层时空特征以生成未来的预测云图序列，从而消除数据采集的时间滞后影响。利用CNN提取预测云图的特征，构建引入时间序列分段与残差连接的改进LSTM网络，建立云图特征到光伏功率的非线性映射模型。实验基于SKIPPD数据集，选取晴天与多云2种典型天气场景进行验证。实验结果表明：在地基云图预测任务中，STFE-GCI模型在图像清晰度与结构相似性指标上均优于ConvLSTM、E3DLSTM等基准模型，展现出更强的时空特征提取能力；在光伏功率预测任务中，STFE-GCI-ILSTM模型在不同天气条件下均表现出更高的预测精度与稳定性，其各项误差指标明显降低，决定系数R²得到有效提升。该方法有效利用了云层状态的时空演化信息，明显提升了超短期光伏功率预测的准确性与鲁棒性。

关键词: 地基云图, 时空特征, 长短期记忆网络, 云图预测, 光伏功率预测

Abstract:

In view of the time lag in ground-based cloud image acquisition and the insufficient accuracy of existing photovoltaic power prediction models, an ultra-short-term photovoltaic power prediction method named spatial-temporal feature enhancement-ground-based cloud image-improved LSTM (STFE-GCI-ILSTM), which is based on spatio-temporal feature enhancement-ground-based cloud images- improved long short-term memory (LSTM) network was proposed. A ground-based cloud image prediction model with multi-scale spatio-temporal feature enhancement (STFE-GCI) was constructed. By spatial feature enhancement, temporal feature enhancement, and multi-scale feature fusion technologies, the deep spatio-temporal features of historical cloud image sequences were extracted to generate future predicted cloud image sequences, thereby eliminating the time lag effect of data acquisition. CNN was used to extract the features of predicted cloud images. An improved LSTM network introducing time series segmentation and residual connections was constructed, and a nonlinear mapping modelfrom cloud image features to photovoltaic power was established. The proposed method was verified based on the SKIPPD dataset under two typical weather scenarios: sunny and cloudy days. The experimental results indicate that in the ground-based cloud image prediction task, the STFE-GCI model outperforms baseline models such as ConvLSTM and E3DLSTM in terms of image clarity and structural similarity, showing a stronger ability of spatio-temporal feature extraction; in the photovoltaic power prediction task, the STFE-GCI-ILSTM model exhibits higher prediction accuracy and stability under different weather conditions; its various error indicators are significantly reduced; the coefficient of determination R² is effectively improved. This method effectively utilizes the spatio-temporal evolution information of cloud states, significantly improving the accuracy and robustness of ultra-short-term photovoltaic power prediction.

Key words: ground-based cloud image, spatio-temporal feature, LSTM network, cloud image prediction, photovoltaic power prediction

中图分类号:

TM615

寇志伟,辛峰悦,崔啸鸣等 . 基于地基云图时空特征增强的超短期光伏功率预测方法[J]. 系统仿真学报, 2026, 38(7): 2068-2081.

Kou Zhiwei,Xin Fengyue,Cui Xiaoming,et al . Ultra-short-term Photovoltaic Power Prediction Method Based on Spatio-temporal Feature Enhancement of Ground-based Cloud Images[J]. Journal of System Simulation, 2026, 38(7): 2068-2081.

图/表 18

图1

图2

图3

图4

图5

图6

图7

图8

表1

图9

表2

表3

图10

图11

表4

图12

图13

表5

参考文献 29

[1]	张智刚, 康重庆. 碳中和目标下构建新型电力系统的挑战与展望[J]. 中国电机工程学报, 2022, 42(8): 2806-2818.
	Zhang Zhigang, Kang Chongqing. Challenges and Prospects for Constructing the New-type Power System Towards a Carbon Neutrality Future[J]. Proceedings of the CSEE, 2022, 42(8): 2806-2818.
[2]	陈涛, 王艳, 纪志成. 基于经验小波变换的光伏功率组合预测模型[J]. 系统仿真学报, 2021, 33(11): 2627-2635.
	Chen Tao, Wang Yan, Ji Zhicheng. Combination Forecasting Model of Photovoltaic Power Based on Empirical Wavelet Transform[J]. Journal of System Simulation, 2021, 33(11): 2627-2635.
[3]	国家能源局. 2024年可再生能源并网运行情况[EB/OL]. (2025-01-27) [2025-06-10]. .
[4]	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.
[5]	朱琼锋, 李家腾, 乔骥, 等. 人工智能技术在新能源功率预测的应用及展望[J]. 中国电机工程学报, 2023, 43(8): 3027-3047.
	Zhu Qiongfeng, Li Jiateng, Qiao Ji, et al. Application and Prospect of Artificial Intelligence Technology in Renewable Energy Forecasting[J]. Proceedings of the CSEE, 2023, 43(8): 3027-3047.
[6]	陈殿昊, 臧海祥, 蒋雨楠, 等. 基于多层次云图特征与宽度学习的超短期光伏功率预测[J]. 电力系统自动化, 2024, 48(22): 131-139.
	Chen Dianhao, Zang Haixiang, Jiang Yunan, et al. Ultra-short-term Photovoltaic Power Forecasting Based on Multi-level Sky Image Features and Broad Learning[J]. Automation of Electric Power Systems, 2024, 48(22): 131-139.
[7]	Pedro H T C, Coimbra C F M, Lauret Philippe. Adaptive Image Features for Intra-hour Solar Forecasts[J]. Journal of Renewable and Sustainable Energy, 2019, 11(3): 036101.
[8]	Niklas Benedikt Blum, Nouri Bijan, Wilbert Stefan, et al. Cloud Height Measurement by a Network of All-sky Imagers[J]. Atmospheric Measurement Techniques, 2021, 14(7): 5199-5224.
[9]	陈志宝, 丁杰, 周海, 等. 地基云图结合径向基函数人工神经网络的光伏功率超短期预测模型[J]. 中国电机工程学报, 2015, 35(3): 561-567.
	Chen Zhibao, Ding Jie, Zhou Hai, et al. A Model of Very Short-term Photovoltaic Power Forecasting Based on Ground-based Cloud Images and RBF Neural Network[J]. Proceedings of the CSEE, 2015, 35(3): 561-567.
[10]	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 Zhiying, Zhou Qingxia, Li Xin, et al. Photovoltaic Power Prediction Based on Image Features of Ground Cloud Image[J]. Proceedings of the CSU-EPSA, 2020, 32(8): 70-76.
[11]	Omaima El Alani, Abraim Mounir, Ghennioui Hicham, et al. Short Term Solar Irradiance Forecasting Using Sky Images Based on a Hybrid CNN-MLP Model[J]. Energy Reports, 2021, 7(S5): 888-900.
[12]	Feng Cong, Zhang Jie. SolarNet: A Sky Image-based Deep Convolutional Neural Network for Intra-hour Solar Forecasting[J]. Solar Energy, 2020, 204: 71-78.
[13]	张臻, 陈天鹏, 王磊, 等. 基于地基云图的超短期太阳辐照预测方法与装置研究[J]. 太阳能学报, 2023, 44(1): 133-140.
	Zhang Zhen, Chen Tianpeng, Wang Lei, et al. Research on Ultra-short-term Solar Irradiance Prediction Method and Device Based on Ground-based Cloud Images[J]. Acta Energiae Solaris Sinica, 2023, 44(1): 133-140.
[14]	辛峰悦, 寇志伟, 刘焱晨, 等. 基于双流特征增强网络的超短期光伏功率预测[J]. 电子测量技术, 2025, 48(5): 65-73.
	Xin Fengyue, Kou Zhiwei, Liu Yanchen, et al. Ultra-short-term Photovoltaic Power Prediction Based on Two-stream Feature Enhancement Network[J]. Electronic Measurement Technology, 2025, 48(5): 65-73.
[15]	梅华威, 杨鹏慧, 余洋. 计及数据漂移改进PatchTST的超短期光伏功率预测[J/OL]. 系统仿真学报. (2025-08-15)[2025-09-02]. .
	Mei Huawei, Yang Penghui, Yu Yang. Ultra-short-term Photovoltaic Power Prediction Based on Improved PatchTST Considering Data Drift[J/OL]. Journal of System Simulation. (2025-08-15)[2025-09-02]. .
[16]	董志强, 郑凌蔚, 苏然, 等. 一种基于IGWO-SNN的光伏出力短期预测方法[J]. 电力系统保护与控制, 2023, 51(1): 131-138.
	Dong Zhiqiang, Zheng Lingwei, Su Ran, et al. An IGWO-SNN-based Method for Short-term Forecast of Photovoltaic Output[J]. Power System Protection and Control, 2023, 51(1): 131-138.
[17]	Wen Haoran, Du Yang, Chen Xiaoyang, et al. Deep Learning Based Multistep Solar Forecasting for PV Ramp-rate Control Using Sky Images[J]. IEEE Transactions on Industrial Informatics, 2021, 17(2): 1397-1406.
[18]	Feng Cong, Zhang Jie, Zhang Wenqi, et al. Convolutional Neural Networks for Intra-hour Solar Forecasting Based on Sky Image Sequences[J]. Applied Energy, 2022, 310: 118438.
[19]	Zhao Xin, Wei Haikun, Wang Hai, et al. 3D-CNN-based Feature Extraction of Ground-based Cloud Images for Direct Normal Irradiance Prediction[J]. Solar Energy, 2019, 181: 510-518.
[20]	柴闵康, 夏飞, 张浩, 等. 基于云图特征自识别的光伏超短期预测模型[J]. 电网技术, 2021, 45(3): 1023-1031.
	Chai Minkang, Xia Fei, Zhang Hao, et al. Ultra-short-term Prediction of Self-identifying Photovoltaic Based on Sky Cloud Chart[J]. Power System Technology, 2021, 45(3): 1023-1031.
[21]	李大海, 王忠华, 王振东. 结合空间域和频域信息的双分支低光照图像增强网络[J]. 计算机应用, 2024, 44(7): 2175-2182.
	Li Dahai, Wang Zhonghua, Wang Zhendong. Dual-branch Low-light Image Enhancement Network Combining Spatial and Frequency Domain Information[J]. Journal of Computer Applications, 2024, 44(7): 2175-2182.
[22]	Tan Mingxing, Pang Ruoming, Le Q V. EfficientDet: Scalable and Efficient Object Detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2020: 10778-10787.
[23]	余光正, 陆柳, 汤波, 等. 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[J]. 中国电机工程学报, 2021, 41(20): 6989-7002.
	Yu Guangzheng, Lu Liu, Tang Bo, et al. An Improved Hybrid Neural Network Ultra-short-term Photovoltaic Power Forecasting Method Based on Cloud Image Feature Extraction[J]. Proceedings of the CSEE, 2021, 41(20): 6989-7002.
[24]	Nie Yuhao, Li Xiatong, Scott A, et al. SKIPP'D: A Sky Images and Photovoltaic Power Generation Dataset for Short-term Solar Forecasting[J]. Solar Energy, 2023, 255: 171-179.
[25]	Shi Xingjian, Chen Zhourong, Wang Hao, et al. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting[C]//Proceedings of the 29th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2015: 802-810.
[26]	Wang Yunbo, Jiang Lu, Yang Ming-Hsuan, et al. Eidetic 3D LSTM: A Model for Video Prediction and Beyond[C]//ICLR 2019. New York: ICLR, 2019: 1-14.
[27]	Wang Yunbo, Wu Haixu, Zhang Ananjin, et al. PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(2): 2208-2225.
[28]	Tan Cheng, Gao Zhangyang, Li Siyuan, et al. SimVPv2: Towards Simple Yet Powerful Spatiotemporal Predictive Learning[J]. IEEE Transactions on Multimedia, 2025, 27: 5170-5184.
[29]	Ruan Guoping, Chen Xiaoyang, Lim E G, et al. On the Use of Sky Images for Intra-hour Solar Forecasting Benchmarking: Comparison of Indirect and Direct Approaches[J]. Solar Energy, 2024, 276: 112649.

参数	设置
多尺度特征通道数	[64, 128, 256, 512]
输入时间步长	8
时间分段数量	2
LSTM隐藏层维度	64
批大小	16
初始学习率	0.001
损失函数	MSE
优化器	Adma

模型	浮点运算数	参数量	MSE	MAE	SSIM	PSNR/dB
ConvLSTM	15.5×10⁶	46.4×10⁹	28.781	293.464	0.910	32.502
E3DLSTM	52.8×10⁶	233.0×10⁹	29.902	295.835	0.908	32.240
PredRNN	24.6×10⁶	94.8×10⁹	28.616	292.401	0.910	32.556
SimVP	43.1×10⁶	14.7×10⁹	29.287	286.796	0.907	33.108
STFE-GCI	17.0×10⁶	15.1×10⁹	28.319	287.815	0.911	33.013

模型	MSE	MAE	SSIM	PSNR/dB
方法I	30.174	318.801	0.906	31.509
方法II	29.864	289.677	0.907	32.998
方法III	30.074	309.833	0.906	31.782
STFE-GCI	28.319	287.815	0.911	33.013

天气	模型	MAE/kW	RMSE/kW	R²
晴天	CNN	0.45	0.61	0.992
	CNN-LSTM	0.47	0.58	0.993
	ConvLSTM	0.53	0.71	0.989
	SCNN	0.53	0.74	0.988
	3DCNN	0.64	0.80	0.986
	STEF-GCI-ILSTM	0.37	0.46	0.995
阴天	CNN	3.26	4.63	0.600
	CNN-LSTM	3.14	4.46	0.630
	ConvLSTM	3.23	4.63	0.601
	SCNN	3.15	4.50	0.622
	3DCNN	3.26	4.64	0.599
	STEF-GCI-ILSTM	3.01	4.30	0.656