Magnetizing inrush current identification using wavelet based gaussian mixture models

The paper presents a novel differential protection approach based on Gaussian Mixture Models (GMM). This method has the advantages of high accuracy and low computational burden. Two common transients such as magnetizing inrush current and internal fault which their mis-identification may lead to mal-operation of differential relays are considered. GMM, the powerful probabilistic pattern classifier is trained with the features extracted by discrete wavelet transform to reduce the computational time of training procedure and enhance the discrimination accuracy. Training data sets are achieved using k-means clustering algorithm. Based on the proposed algorithm, a high speed differential relaying (a quarter of a cycle) without dependency on a specific threshold is performed. The suitable performance of this method is demonstrated by simulation of different faults and switching conditions on a power transformer in PSCAD/EMTDC software. Sympathetic and recovery inrush currents were also simulated and investigated. The proposed algorithm is also evaluated by the data collected from a prototype laboratory power transformer. It provides a high operating sensitivity for internal faults and remains stable for inrush currents even in noisy conditions. Since the discrimination method is done with stochastic characteristics of signals without application of any deterministic index, more reliable and accurate classification is achieved.