Incremental segmented slope residential load pattern clustering based on three-stage curve profiles

ABSTRACTThis paper tackles high computational complexity in using Euclidean distance for residential load profiles (RLPs) similarity by proposing a three-stage incremental segmented slope clustering framework. The first two stages involve static clustering, where we obtain typical residential load profiles through piecewise slope clustering. In the third stage, dynamic clustering is performed based on the slope similarity of RLPs. This method enhances clustering performance and reduces computation cost, outperforming various benchmarks, with simulation results confirming the framework's effectiveness.KEYWORDS: Residential load profilesthree-stage segmented slope clusteringincremental pattern clusteringslope similarity Nomenclature σ(dia,dib)=a metric that evaluates if the slope aspect of xit on day a and day b at time j is identical or notadvd=the average deviation of Ud from the minimum value of each columnadvs=the average deviation of Us from the minimum value of each columncir=the r-th clustering centre obtained after clustering user i in the first stagecci=the final TRLPs of all customers in the static data setdijt=the slope direction of the j-th segment of xitdxj=the deviation between Gj and its maximum value max(Gj)fd,j=the deviation of an element in Ujd from its minimum value min(Ujd)fs,j=the deviation of an element in Ujs from its minimum value min(Ujs)G=segmented slope co-directional matrixg(dia,dib)=the number of slope segments with the same direction on day a and day b of xitgij=the number of segmented slopes in the same direction on the i-th and j-th days of the userki=the number of categories after the first stage clusteringpijt=the slope steepness of the j-th segment of xituaqd=the average slope difference between the RLP a and the clustering centre q in different slope direction sectionuaqs=the average slope difference between the RLP a and the clustering centre q in the same slope direction sectionUjd=the average slope dissimilarity between other t curves and jUjs=the average slope similarity between other t curves and jxit=the RLPs of user i on the t-th dayz=the number of current TRLPsDBI=Davidson-Boding IndexISSC=Incremental Segmented Slope ClusteringRLP=Residential Load ProfileSOM=Self-Organizing MapSSC=Segmented Slope ClusteringTRLP=Typical Residential Load ProfileWSOM=Weighted Self-Organizing MapDisclosure statementNo potential conflict of interest was reported by the authors.Additional informationFundingThis work was supported by the China Southern Power Grid Company Limited under the Grant No. 036000KK52222009 (GDKJXM20222125).