基于原型提取和聚类的光伏电站快速集群划分方法

A method for rapid cluster partitioning of photovoltaic plants based on prototype ex⁃traction and clustering

在光伏发电渗透率不断提高的背景下,针对光伏电站集群划分效果差、耗时长的问题,提出一种基于原型提取和聚类的光伏电站快速集群划分方法。首先,对光伏数据进行预处理,消除不同数据在量级与量纲上的差异性;然后,基于Pearson相关系数法筛选出对光伏出力影响较大的因素,然后通过设置随机抽样、k-means++和改进谱聚类3个环节,分别实现光伏电站的抽样、原型提取和原型聚类;继而基于枚举法和分层优化的思想,搜索上述环节的最优超参数;最后,设置不同场景进行算例对照,计算聚类内外指标和聚类时间指标,通过综合分析,验证了所提方法在解决大规模光伏电站快速聚类问题上的有效性。

The penetration rate of photovoltaic power generation keeps increasing.To address the issues of poor clus-ter partitioning and lengthy processing times for photovoltaic power station clusters,the paper proposes a method for rapid cluster partitioning method for photovoltaic(PV)plants based on prototype extraction and clustering.Firstly,photovoltaic data is preprocessed to eliminate differences in magnitude and dimensionality among data sets.Subse-quently,influential factors on photovoltaic output power are identified using the Pearson correlation coefficient method.Random sampling,k-means++,and an improved spectral clustering method are then employed for sam-pling,prototype extraction,and prototype clustering of PV plants,respectively.Building upon an enumeration ap-proach and hierarchical optimization,optimal hyperparameters for the aforementioned processes are determined.Fi-nally,various scenarios are set up for case study comparisons,calculating both intra-cluster and inter-cluster indi-cators as well as clustering time metrics.Through comprehensive analysis,the effectiveness of the proposed method in addressing the rapid clustering for large-scale PV plants is validated.