| Accurately grasping the power load and time domain characteristics of power users plays an important role in improving the quality of power supply service,optimizing power grid dispatching,reducing waste of resources and improving economic benefits.Based on big data technology,this paper studies an improved fast density peak clustering algorithm,which is applied to the analysis and processing of power load data of large industrial users,and achieved the desired result.The original fast density peak clustering algorithm can not directly analyze the data in time domain.Before clustering,we need to manually specify the parameters truncation distance,and we need judgement when selecting the cluster center.In this paper,the discrete wavelet transform(DWT)is used to decompose the data and obtained the spectrum components on different time scales,and then the load characteristics are analyzed in time domain.Combined with the idea of distance based KNN,we redefine the local density and the distance between adjacent density points in the original algorithm.When we determine the cluster center,we use the clustering center automatic selection strategy to avoid manual selection.After reconstruct the cluster center of each spectral component,we get the typical load characteristic curve and get the ideal clustering result.This paper adopts the improved clustering algorithm and the traditional clustering algorithms cluster the load data of industrial users.It also evaluates the error squared sum of industrial users clustering results and the dispersion and compactness of clusters,Compared with the advantages of the traditional algorithm,the improved algorithm can reduce the human intervention and analyze the load characteristics more accurately in the time domain. |
