Find in Library

A polar projection-based algorithm is proposed to reduce the computational complexity associated with dimension reduction in unsupervised learning. This algorithm employs K-means clustering. A new distance metric is developed to account for peak consumption in cluster consumer load profiles. It is used to cluster the load profiles according to both total and peak consumption. To accelerate the clustering process, a stochastic-based approach is developed to reduce the search space to find the cluster centers. Numerical results are presented which show a significant reduction in computational complexity using both polar-based and stochastic-based clustering compared to conventional approaches. Further, the estimation error is low.