Classification of sags according to their origin based on the waveform similarity J. Melendez 1 , X. Berjaga 1 , S. Herraiz 1 , V. Barrera 1 , J. Sanchez 2 and M. Castro 2 Abstract- A statistical method for classification of sags according to their origin downstream or upstream from the recording point is proposed in this work. The goal is to obtain a statistical model using the sag waveforms useful to characterise one type of sags and to discriminate them from the other type. This model is built on the basis of multi-way principal component analysis an later used to project the available registers in a new space with lower dimension. Thus, a case base of diagnosed sags is built in the projection space. Finally classification is done by comparing new sags against the existing in the case base. Similarity is defined in the projection space using a combination of distances to recover the nearest neigbours to the new sag. Finally the method assigns the origin of the new sag according to the origin of their neighbours. Index Terms- Fault location, voltage sag (dip), pattern classification, Power quality monitoring, Principal Component Analysis. I. INTRODUCTION U TILITY companies have increased the number of power quality monitors installed in the distribution substations and are very interested in developing reliable methods to efficiently exploit the information contained in these registers. The goal in this work has been focused on the discrimination between sags originating in the transmission (RV) and distribution (MV) networks. With this aim, sags registered in three 25kV distribution substations have been used as case base. Additionally, the utility has provided information related to the origin, upstream (RV) or downstream (MV) from the transformer, of them. Data mining principles can be applied to obtain the desired information and manage the huge volume of data contained in these registers efficiently. The basic principles of these strategies involve automatic classification, clustering, or I This research has been made possible by the interest and participation of ENDESA DISTRIBUCION and its Power Quality Department. It has also been supported by the research project DPI2006-09370 funded by the Spanish Government. The authors (1) are with the eXiT Group in the Institute of Informatics and Applications of the University of Girona, Spain, Girona, Campus Montilivi, 17071, e-mail: xberiaga@eia.udg.edu, sherraiz(lfeia.udg.es, vbarrera({i)eia. udg.edu. Authors (2) are with the PQ Department of Endesa Distribuci6n, Barcelona, (Spain), e-mail The eXiT is part of Automation Engineering and Distributed Systems (AEDS) research group, awarded with a consolidated distinction (SGR- 00296) for the 2005-2008 period in the Consolidated Research Group (SGR) project of the Generalitat de Catalunya. 978-1-4244-2218-0/08/$25.00 ©2008 IEEE. pattern matching to recognize disturbances according to similarity criteria and associate them with the most plausible causes and origins. Researchers have classified sags according to their origins to assist utilities in locating faults. Determining whether sags have occurred in the distribution or transmission networks precedes the localization and mitigation stages [1]. Typical classification according to the origin consists in discriminating between transmission (or high voltage) and distribution (or medium voltage) origins. For this purpose, phase analysis and an unsupervised method were compared in [2] by extracting some temporal descriptors from the RMS representation of sags and using a Learning Algorithm for Multivariate Data Analysis (LAMDA). Recent research has also identified similarities among sags using the variability in the information contained in the waveform in statistical analyses based on Principal Component Analysis (PCA), which allows dimensionality reduction before similarity criteria are applied to sags, assigning them to different classes. In [2] sags are categorized into three classes using certain features run through a fuzzy system. A more recent method for locating the origin of a voltage sags in a power distribution system using the polarity of the real current component relative to the monitoring point has been introduced in [1]. Other approaches proposed for classifying voltage sags are related to defining and describing sag types with regard to their general three-phase nature. With these approaches, sags can be divided up according to the number of sagged phases and the presence of asymmetries using either the magnitude or the angle between phasors to identify sag typologies. Other strategies are related to evaluate both the minimum magnitude and the total duration of sags. This group of classifiers eliminates any possibility of classifying sags using their three- phase nature. With this approach, the sags are reduced to one simple square shape sag, which is represented by the minimum of all RMS phase voltages during the sag and the total duration of the sag in all sagged phases. Other sag, classification strategies take advantage of attributes extracted from the RMS waveform to represent sags in a feature space where classification algorithms are applied ([2]-[4D. In this paper we present new results obtained with a classification method based on the definition of similarity criterion in the projection space obtained when the Principal Component Analysis (PCA) is applied to sags waveforms [14]. The method proposes the exploitation of the whole information contained in the voltage and current waveforms instead of Authorized licensed use limited to: UNIVERSITAT DE GIRONA. Downloaded on April 27,2010 at 07:26:33 UTC from IEEE Xplore. Restrictions apply.