A probabilistic load modelling approach using clustering algorithms

In this paper, a novel probabilistic load modeling approach is presented. The proposed approach starts by grouping the 24 data points representing the hourly loading of each day in one data segment. The resulting 365 data segments representing the whole year loading profile are evaluated for similarities using principle component analysis; then segments with similar principal components are grouped together into one cluster using clustering algorithms. For each cluster a representative segment is selected and its probability of occurrence is computed. The results of the proposed algorithm can be used in different studies to model the long term behavior of electrical loads taking into account their temporal variations. This feature is possible as the selected representative segments cover the whole year. The designated representative segments are assigned probabilistic indices that correspond to their frequency of occurrence, thus preserving the stochastic nature of electrical loads.