A voltage dip is a short duration reduction in voltage magnitude due to a short duration increase in current magnitude. Causes of dips are, among others, electrical faults, large motor starting, transformer energizing and failure of power-electronic converters.
Voltage dips are considered as a very important power quality issue because they lead to trip or malfunction of sensitive loads especially in industrial process installations and subsequently they lead to high costs.
In this thesis the overall aim is extracting additional information from large voltage dip monitoring databases. An important step to this end is providing efficient characterization methods for voltage dips. Voltage dip characterization aids by describing voltage dip events (a set of voltage waveforms with high time resolution) as a limited number of values such that this set gives as much as possible information about the dip. This thesis contributes to the voltage dip characterization development through three different methods.
The first method consists of a systematic way for comparison different sets of voltage dip characteristic. With this method, both real-measured and synthetic voltage dips are applied to generic models of sensitive loads. The best set of characteristics, for representing the voltage dip, is the one best enables the reproduction of the behaviour of equipment when exposed to real-measured voltage dips.
The second method compares 12 different sets of characteristics for describing three-phase single-events.. The method determines the most efficient and feasible way that gives more realistic characteristics as well as comparable with existing standard methods. The proposed set of characteristics has been proposed for inclusion in international standard documents.
The third method enables the extraction of dip characteristics based on machine learning approaches. It is applicable for characterization of multi-stage voltage dips in particular and for single-stage (normal) voltage dips as well. The proposed method uses the space-phasor model of three-phase voltages as an input data for k-means clustering algorithm. Then the calculated data are modeled as a general form of an ellipse by exploiting logistic regression algorithm. Finally the optimized obtained ellipse parameters are applied to calculate single-segment characteristics for each individual stage of a multi-stage voltage dip.
Further, all proposed methods are implemented in an Matlab environment and validated by applying them to a large number of real-measured voltage dips in actual HV and MV power networks and some suitable synthetic voltage dips.