Distribution Power System Fault Detection, Location, And Classification Using Artificial Neural Network

Abstract

Distribution power system fault detection, classification, and localization are critical for a power system's protection and maintenance. Those methods rely on current and voltage transformer measurements of electrical quantities. The purpose of this thesis is to address a reliable and high-quality power supply in the distribution system before a distribution system fault occurs. Additionally, precautions could be taken to assist the utility in quickly and cost-effectively resolving issues of faults. The major problem in the study area is that the power delivered from distribution systems to the end-user level is not reliable due to different types of faults on distribution feeder lines. To overcome such problems the detection, classification, and location of faults using Artificial Neural Networks (ANN) is the focus of this thesis. This is accomplished by employing the back-propagation neural network (BPNN) with three-phase rms and sequence voltages and currents as inputs respectively. MATLAB will be used to carry out the simulations. The results of analyzing ANN with one hidden layer validate the method's usefulness and a variety of ANN configurations, some of which are not shown here, are built to obtain well trained neural networks, and six ANN configurations (two of them for fault detection, the left two are for classification and two of them are for fault location) are shown to illustrate various ANN configurations for fault detection, classification, and location. From training, testing, and validation different linear regression, error histogram, and validation MSE performance were determined. 650, 750, and 1000, data were generated for fault detection, classification, and location respectively. Fault location using percentage error is discussed from actual and estimated distance location for different fault resistance. Finally, the levenberg marquardt algorithm and mean square error to evaluate the performance of the detector/classifier, as well as fault locator, were used. The results show that the validation performance for the fault detector is 3.6456𝑒 −10 , the classifier is 1.8326𝑒 −10 and for fault, the locator is 1.1809𝑒 −11 . The system can detect if there is a fault or not, can classify the different fault types, and pinpointing the fault location very precisely.