基于DGAT-Transformer组合算法的区域光伏出力数据质量增强与超短期功率预测方法

任惠; 于光发; 强涵悦; 王飞; 甄钊; 常喜强

doi:10.19912/j.0254-0096.tynxb.2024-2352

PDF(2229 KB)

太阳能学报 ›› 2026, Vol. 47 ›› Issue (5) : 639-649. DOI: 10.19912/j.0254-0096.tynxb.2024-2352

基于DGAT-Transformer组合算法的区域光伏出力数据质量增强与超短期功率预测方法

任惠¹, 于光发¹, 强涵悦¹, 王飞¹, 甄钊¹, 常喜强²

作者信息 +

REGIONAL PHOTOVOLTAIC OUTPUT DATA QUALITY ENHANCEMENT AND ULTRA-SHORT-TERM POWER PREDICTION METHOD BASED ON DGAT-TRANSFORMER ENSEMBLE ALGORITHM

Ren Hui¹, Yu Guangfa¹, Qiang Hanyue¹, Wang Fei¹, Zhen Zhao¹, Chang Xiqiang²

Author information +

文章历史 +

摘要

针对区域内各电站的光伏数据存在不同程度的缺失值而导致预测精度低的问题,提出基于时空关联性与动态图注意力网络建模的光伏功率缺失数据填充和基于周期—趋势分量解耦模型的分布式光伏功率预测的新方法框架。首先基于贝叶斯算法优化后的变分模态分解模型对光伏功率分解为多个趋势平稳的子序列,并将对应子序列合并构造特征矩阵。其次,提出基于动态图注意力网络和Transformer组合模型,提高对特征矩阵在时空双维度上动态关联关系表征的准确性,并利用此模型对缺失时刻各个子序列预测后加以重构,得到光伏电站的完备数据集。最后,引入基于变分推断原理的周期—趋势分量解耦的预测模型（LaST）,将光伏功率分解为周期性分量与趋势性分量,分别预测后加以重构,以提高光伏预测精度。仿真结果验证了该文所提方法获得的数据集真实性更高,并以此作为训练样本的预测精度显著优于其他方法,同时验证了LaST模型相较于传统算法在光伏预测中的优越性能。

Abstract

To address the decline in prediction accuracy caused by varying degrees of missing photovoltaic data across regional power stations, this paper proposes a novel framework that integrates photovoltaic power data imputation based on spatiotemporal correlation and dynamic graph attention network modeling with distributed photovoltaic power prediction leveraging a period-trend component decoupling approach. Specifically, a variational mode decomposition （VMD） model optimized via Bayesian algorithms is first employed to decompose the photovoltaic power series into multiple trend-stationary subsequences, which are then merged to form a comprehensive feature matrix. Subsequently, a hybrid model combining a dynamic graph attention network and Transformer is developed to more accurately capture the dynamic correlations within the feature matrix across spatiotemporal dimensions. This model is utilized to predict and reconstruct the missing subsequences, resulting in a complete photovoltaic dataset. Finally, a prediction model based on the principle of variational inference, termed the period-trend component decoupling model （LaST）, is introduced to further decompose the photovoltaic power into periodic and trend components. These components are individually modeled and predicted before being reconstructed to enhance overall prediction accuracy. Simulation results validate that the proposed method generates datasets with higher fidelity, which, when used as training samples, significantly improve prediction performance. Moreover, the superiority of the LaST model over conventional approaches in photovoltaic power forecasting is also demonstrated.

导出引用

任惠, 于光发, 强涵悦, 王飞, 甄钊, 常喜强. 基于DGAT-Transformer组合算法的区域光伏出力数据质量增强与超短期功率预测方法[J]. 太阳能学报. 2026, 47(5): 639-649 https://doi.org/10.19912/j.0254-0096.tynxb.2024-2352

Ren Hui, Yu Guangfa, Qiang Hanyue, Wang Fei, Zhen Zhao, Chang Xiqiang. REGIONAL PHOTOVOLTAIC OUTPUT DATA QUALITY ENHANCEMENT AND ULTRA-SHORT-TERM POWER PREDICTION METHOD BASED ON DGAT-TRANSFORMER ENSEMBLE ALGORITHM[J]. Acta Energiae Solaris Sinica. 2026, 47(5): 639-649 https://doi.org/10.19912/j.0254-0096.tynxb.2024-2352

中图分类号： TM615

参考文献

[1] 阮呈隆, 李康平, 李正辉, 等. 分布式光伏集群功率短期预测的空间互补特性初探[J]. 电力系统自动化, 2024, 48(3): 42-50.
RUAN C L, LI K P, LI Z H, et al.Preliminary study on spatial complementarity characteristics of short-term power prediction for distributed photovoltaic clusters[J]. Automation of electric power systems, 2024, 48(3): 42-50.
[2] 王蒙, 张文朝, 汪莹, 等. 高比例光伏接入的电力系统暂态过电压控制策略[J]. 太阳能学报, 2023, 44(10): 148-155.
WANG M, ZHANG W C, WANG Y, et al.Transient overvoltage control strategy of power system considering high proportion photovoltaic access[J]. Acta energiae solaris sinica, 2023, 44(10): 148-155.
[3] DAI Y M, YU W J, LENG M M.A hybrid ensemble optimized BiGRU method for short-term photovoltaic generation forecasting[J]. Energy, 2024, 299: 131458.
[4] 王玉庆, 徐飞, 刘志坚, 等. 基于动态关联表征与图网络建模的分布式光伏超短期功率预测[J]. 电力系统自动化, 2023, 47(20): 72-82.
WANG Y Q, XU F, LIU Z J, et al.Ultra-short-term power forecasting of distributed photovoltaic based on dynamic correlation characterization and graph network modeling[J]. Automation of electric power systems, 2023, 47(20): 72-82.
[5] FENG H F, YU C S.A novel hybrid model for short-term prediction of PV power based on KS-CEEMDAN-SE-LSTM[J]. Renewable energy focus, 2023, 47: 100497.
[6] ZHANG J J, HONG L Q, IBRAHIM S N, et al.Short-term prediction of behind-the-meter PV power based on attention-LSTM and transfer learning[J]. IET renewable power generation, 2024, 18(3): 321-330.
[7] 王福忠, 王帅峰, 张丽. 基于VMD-LSTM与误差补偿的光伏发电超短期功率预测[J]. 太阳能学报, 2022, 43(8): 96-103.
WANG F Z, WANG S F, ZHANG L.Ultra short term power prediction of photovoltaic power generation based on VMD-LSTM and error compensation[J]. Acta energiae solaris sinica, 2022, 43(8): 96-103.
[8] LI G Z, DING C J, ZHAO N N, et al.Research on a novel photovoltaic power forecasting model based on parallel long and short-term time series network[J]. Energy, 2024, 293: 130621.
[9] CHEN Y, LI X L, ZHAO S G.A novel photovoltaic power prediction method based on a long short-term memory network optimized by an improved sparrow search algorithm[J]. Electronics, 2024, 13(5): 993.
[10] 王晓霞, 俞敏, 霍泽健, 等. 基于近邻传播聚类与LSTNet的分布式光伏电站群短期功率预测[J]. 电力系统自动化, 2023, 47(6): 133-141.
WANG X X, YU M, HUO Z J, et al.Short-term power forecasting of distributed photovoltaic station clusters based on affinity propagation clustering and long short-term time-series network[J]. Automation of electric power systems, 2023, 47(6): 133-141.
[11] HERRERA CASANOVA R, CONDE A.Enhancement of LSTM models based on data pre-processing and optimization of Bayesian hyperparameters for day-ahead photovoltaic generation prediction[J]. Computers and electrical engineering, 2024, 116: 109162.
[12] 余晓霞, 汤宝平, 王伟影, 等. 复杂工况条件下多头注意力双向长短时记忆网络的风电机组缺失数据修复方法研究[J]. 机械工程学报, 2023, 59(14): 1-9.
YU X X, TANG B P, WANG W Y, et al.Repairing deteriorated data of wind turbines by multi-head attention bi-directional long short time memory networks under complex working conditions[J]. Journal of mechanical engineering, 2023, 59(14): 1-9.
[13] 毛嘉铭, 刘光宇, 罗凯元. 结合数据增强及组合算法的短期光伏功率预测[J]. 电力系统及其自动化学报, 2024, 36(8): 133-141.
MAO J M, LIU G Y, LUO K Y.Short-term photovoltaic power prediction using data augmentation and combined algorithms[J]. Proceedings of the CSU-EPSA, 2024, 36(8): 133-141.
[14] ZHONG S P, WANG X M, XU B, et al.Hybrid network model based on data enhancement for short-term power prediction of new PV plants[J]. Journal of modern power systems and clean energy, 2024, 12(1): 77-88.
[15] 王剑斌, 傅金波, 陈博. 基于强化学习的多模型融合光伏发电功率预测方法[J]. 太阳能学报, 2024, 45(6): 382-388.
WANG J B, FU J B, CHEN B.Multi-model fusion photovoltaic power generation prediction method based on reinforcement learning[J]. Acta energiae solaris sinica, 2024, 45(6): 382-388.
[16] 尹晓敏, 孟祥剑, 侯昆明, 等. 一种计及空间相关性的光伏电站历史出力数据的修正方法[J]. 山东大学学报(工学版), 2021, 51(4): 118-123.
YIN X M, MENG X J, HOU K M, et al.Correction method for historical output data of photovoltaic power plant considering spatial correlation based on artificial neural network[J]. Journal of Shandong University (engineering science), 2021, 51(4): 118-123.
[17] 乔颖, 孙荣富, 丁然, 等. 基于数据增强的分布式光伏电站群短期功率预测(一): 方法框架与数据增强[J]. 电网技术, 2021, 45(5): 1799-1808.
QIAO Y, SUN R F, DING R, et al.Distributed photovoltaic station cluster gridding short-term power forecasting part Ⅰ: methodology and data augmentation[J]. Power system technology, 2021, 45(5): 1799-1808.
[18] LAI W Z, ZHEN Z, WANG F, et al.Sub-region division based short-term regional distributed PV power forecasting method considering spatio-temporal correlations[J]. Energy, 2024, 288: 129716.
[19] 杨晶显, 张帅, 刘继春, 等. 基于VMD和双重注意力机制LSTM的短期光伏功率预测[J]. 电力系统自动化, 2021, 45(3): 174-182.
YANG J X, ZHANG S, LIU J C, et al.Short-term photovoltaic power prediction based on variational mode decomposition and long shortterm memory with dual-stage attention mechanism[J]. Automation of electric power systems, 2021, 45(3): 174-182.
[20] WANG G, JIA R, LIU J H, et al.A hybrid wind power forecasting approach based on Bayesian model averaging and ensemble learning[J]. Renewable energy, 2020, 145: 2426-2434.
[21] 董俊, 刘瑞, 束洪春, 等. 基于BIRCH聚类的L-Transformer分布式光伏短期发电功率预测[J]. 高电压技术, 2024, 50(9): 3883-3893.
DONG J, LIU R, SHU H C, et al.Short-term distributed photovoltaic power generation prediction based on BIRCH clustering and L-Transformer[J]. High voltage engineering, 2024, 50(9): 3883-3893.
[22] 黄冬梅, 陈欢, 王宁, 等. 基于自适应图注意力网络的短期用户负荷预测[J]. 电力系统保护与控制, 2023, 51(20): 140-149.
HUANG D M, CHEN H, WANG N, et al.Short-term user load prediction based on an adaptive graph attention network[J]. Power system protection and control, 2023, 51(20): 140-149.
[23] 王维高, 魏云冰, 滕旭东. 基于VMD-SSA-LSSVM的短期风电预测[J]. 太阳能学报, 2023, 44(3): 204-211.
WANG W G, WEI Y B, TENG X D.Short-term wind power forecasting based on VMD-SSA-LSSVM[J]. Acta energiae solaris sinica, 2023, 44(3): 204-211.
[24] 黄旭锐, 于丰源, 杨波, 等. 基于Transformer网络和多任务学习的园区综合能源系统电-热短期负荷预测方法[J]. 南方电网技术, 2023, 17(1): 152-160.
HUANG X R, YU F Y, YANG B, et al.Short-term electric-thermal load forecasting method for park-level integrated energy system based on transformer network and multi-task learning[J]. Southern power system technology, 2023, 17(1): 152-160.