•  
  •  
 

Keywords

similarity; KPCA-OPTICS; clustering; quantile regression; power anomaly perception

Abstract

In response to the lack of professional monitoring and difficulty in accurately locating abnormal sites in distributed photovoltaic power stations, a distributed photovoltaic power anomaly perception method is proposed based on kernel principal component analysis-ordering points to identify the clustering structure (KPCA-OPTICS) clustering, with the help of the similarity and correlation of nearby distributed photovoltaic power station output. Firstly, based on the output data of photovoltaic power stations, the OPTICS clustering algorithm is used to cluster multiple power stations. The KPCA is then employed to perform dimensionality reduction on the clustering data to lower the influence of high-dimensional data on the clustering accuracy of the OPTICS algorithm. Secondly, the anomaly perception processing is carried out with the divided clusters as the target. The output of the clusters under different weather conditions is weighted equally to gain the output curve characterizing the overall output of the clusters. The output interval of the clusters is fitted by quantile regression (QR) and serves as the anomaly perception basis for the distributed photovoltaic (DPV) clusters. At last, the distributed photovoltaic power data set in a certain city in southern China is applied as the actual verification data for the simulation experiment. The results show that the method can effectively perceive power anomalies in distributed photovoltaic systems, with a high detection rate and precision and a low false alarm rate, and has good model scalability for practical deployment.

DOI

10.19781/j.issn.1673-9140.2026.01.017

First Page

174

Last Page

184

Download

Share

COinS