Least-squares versus LMS parametric approaches for power quality events segmentation

Abstract

Power quality monitoring requires knowing when the start of the perturbation takes place, and also when it ends; in this way, the voltage or current signals are divided into segments. In this work, we follow previously developed ideas in the literature and resort to parametric modelling to achieve the perturbed signal segmentation. What we propose here is the use of adaptive AR modelling identification, in particular Recursive Least Squares and Least Mean Squares, as opposed to a block-based approach used elsewhere. Overdetermined systems, both block-wise and adaptively are also included among the analysed methods. Simulations show that although being computationally lighter, and hence more suitable to real-time implementations, segments limits are accurately located by adaptive algorithms most of the cases.