计及NWP信息缺失的数据共享与GRA权重优化的分布式光伏电站功率预测

杨锡运; 杨岩; 孟令卓超; 彭琰; 王晨旭

杨锡运,杨岩,孟令卓超,彭琰,王晨旭.计及NWP信息缺失的数据共享与GRA权重优化的分布式光伏电站功率预测[J].电测与仪表,2025,62(4):172-179.
YANG Xiyun,YANG Yan,MENG Lingzhuochao,PENG Yan,WANG Chenxu.Data sharing and GRA weight optimization for power prediction of distributed photovoltaic power plant considering missing NWP information[J].Electrical Measurement & Instrumentation,2025,62(4):172-179.

计及NWP信息缺失的数据共享与GRA权重优化的分布式光伏电站功率预测

Data sharing and GRA weight optimization for power prediction of distributed photovoltaic power plant considering missing NWP information

DOI：10.19753/j.issn1001-1390.2025.04.021

中文关键词: 分布式光伏电站，空间相关性，数据共享，权重优化，一维卷积神经网络

英文关键词:distributed photovoltaics, spatial correlation, data sharing, weight optimization, one-dimensional convolutional neural network

基金项目:国网浙江省电力有限公司科技项目（5211DS220009）

作者中文名	作者英文名	单位
杨锡运	YANG Xiyun	华北电力大学控制与计算机工程学院
杨岩	YANG Yan	华北电力华北电力大学控制与计算机工程学院大学
孟令卓超	MENG Lingzhuochao	华北电力大学控制与计算机工程学院
彭琰	PENG Yan	国网浙江省电力有限公司电力科学研究院
王晨旭	WANG Chenxu	国网浙江省电力有限公司电力科学研究院

摘要点击次数: 50

中文摘要:

由于光伏发电的出力具有很强的间歇性和波动性，大规模光伏电站的接入会冲击电网的稳定性，因此对光伏出力进行精准预测至关重要。此外，由于部分光伏电站无法获得用于功率预测的相关数值天气预报（numerical weather prediction，NWP）信息，这对电网的安全稳定运行又提出了新的挑战。基于此，文中提出一种基于数据共享和灰色关联度分析（grey relation analysis，GRA）权重优化的分布式光伏电站功率预测模型。利用K-means算法对光伏电站进行出力空间相关性聚类，构建多电站数据共享集群，通过相似日数据筛选和BP(back propagation)神经网络神经网络对单个参考电站进行出力预测，利用GRA对参考电站进行权重优化，并通过一维卷积神经网络（1D convolutional neural network，1DCNN）对缺失NWP数据的目标电站出力进行预测。以河北省部分市十个分布式光伏电站进行算例分析，结果表明晴天预测的均方根误差为3.34%，非晴天预测的均方根误差为9.15%，具有较高的准确性和可行性，为电网的稳定运行奠定了基础。

英文摘要:

Since the output of photovoltaic power generation has a strong intermittency and volatility, the access of large-scale photovoltaic power plants will impact the stability of the power grid, so it is crucial to accurately predict the output of photovoltaics. In addition, because some photovoltaic power plants cannot obtain the relevant numerical weather prediction (NWP) information for power prediction, this poses new challenges to the safe and stable operation of the power grid. On this basis, this paper proposes a power prediction model for distributed photovoltaic power plant based on data sharing and grey relation analysis (GRA) weight optimization. Firstly, the K-means algorithm is used to cluster the output spatial correlation of photovoltaic power plants, and GRA is used to optimize the weight of the reference power station, and the output of the target power station with missing NWP data is predicted by one-dimensional convolutional neural network (1DCNN). According to the example analysis of ten distributed photovoltaic power plants in some cities in Hebei Province, the results show that the RMSE of sunny day prediction is 3.34%, and the RMSE of non-sunny day prediction is 9.15%, which has high accuracy and feasibility, and lays a foundation for the stable operation of the power grid.

查看全文查看/发表评论下载PDF阅读器

关闭