Power Load Estimation in Smart Grids via k-Means Clustering using Sensor Networks

Abstract

In this paper, estimating real and reactive power measurements provided in smart grids through wireless sensor networks is considered. The communication channel is assumed to suffer from additive white Gaussian noise (AWGN). k-means clustering is used to learn the underlying structure of the collected power measurements. Then, nearest-neighbour method is used to estimate the power measurements from the noisy received measurements. Two clustering approaches are proposed. First, clustering the real and reactive power measurements individually. Second, combining the power measurements and clustering jointly. Simulation results show very small estimation errors for both methods even if a small number of clusters is small, where, the individual clustering performs better. On the other hand, the joint clustering method performs better if the number of clusters increases.