| المستخلص: |
In order to meet the power quality requirements-especially with an existence of power electronic devices, nonlinear loads, microprocessors and distributed generators (DGs)- which have appeared in modern power systems, the increased concern about Power Quality (PQ) problems – such as system harmonics and voltage variations - from electric utilities and customers takes place and becomes an important field of study due to their impacts on sensitive equipments as well as the bad consequences on the other competitive and economic issues. Power quality can be defined as any power problem encountered in voltage, current, or frequency which can affect customer devices and resulting in malfunction in their equipments. Voltage quality is the main concern of the power quality studies that the most of PQ problems are caused by voltage variations. This last general term of 'voltage variations' is used to describe any voltage distortion including sag, swell or interruption of the supply voltage. In this thesis, detection and classification of short time voltage variations -sag, swell and interruption with their time duration will be held using artificial neural network (ANN) based on voltage envelope detection technique extracted from Hilbert Transform (HT). Hilbert Transform is used to extract the envelope of the voltage signal as this envelope can be used to detect any type of voltage variations and also to be an input to the neural network. Classification of voltage variations is performed using an artificial neural network consisting of input, hidden and output layers .One input layer will be used with ten neuron receiving its input from the voltage envelope. The output layer has four states: -1 (for sag) or 0 (for interruption) or 1 (for swell) or 0.5 (for normal state).Training of ANN is performed using neural network algorithm with tan-sigmoid transfer function. Computer simulation is used for training and testing of the voltage envelope patterns using Matlab/Simulink. The proposed envelop based classifier results show an accuracy of 99.225% with a Mean Square Error (MSE) of 0.0096 when a 30 different levels of voltage sag, swell, and interruption of the 16000 samples are trained and simulated. In addition, 3 different levels of each type of variations with different time durations are correctly detected and classified with an average accuracy of 98.57%. The average classification accuracy of voltage sag, swell, and interruption are 98.1%, 98.5%, and 99.2% respectively. Both the best and the worst classification accuracy are 99.4%, and 96.85% respectively. A simulation for single line to ground fault (SLG) is also performed showing the capability of the proposed classifier of detecting both beginning and ending time of each piece of variation in an efficient way with an average accuracy of 98.95%.
|
