Seyyed Meysam Seyyed Barzegar

  Abstract

  One of the challenging protection problems is distribution network islanding since may cause serious problems for network equipment specially distributed generators. As most basic treatment to protect networks facing islanding, islanding detection has been discussed in this dissertation in depth. Used method to do that is passive one. Passive method is characterized by measuring detection parameters at distributed generators terminals. Main used detection parameter is negative sequence voltage component. Regarding various network fluctuations occurring at different situations in terms of generation and consumption equilibrium, the accuracy of detection method may be impacted. Therefore, beside negative sequence voltage component employing synchronous generator field voltage, positive sequence impedance, voltage harmonics distortion factor and defining indices based on voltage wavelet transform output have been considered in order to distinguishing condition caused by islanding from other six network changes. Connect and disconnect loads and capacitor banks, symmetric and asymmetric faults, voltage fluctuations, and voltage notch are those changes which have been distinguished from network islanding in this dissertation. Since passive methods treat based on determining threshold limit for detection parameters, inaccurate determining of these limits may lead to incorrect results. As r result, pattern recognition methods have employed to overcome this problem and omitting threshold limit determining step. Use of Support Vector Machine (SVM) and Neural Network (NN) to classify trained data in island and non-island classes, have demonstrate the efficiency of abovementioned methods to detect island situation. Simulation results show that proper characteristic selection has a major effect on data classification capability.

  Date: 22 September 2010 Time:13

  Place: EE Department Seminar Room