The majority of electric faults are ground-faults. The effect of a single phase to ground-fault must be minimized. The ability to detect and classify the type of fault plays a great role in the protection of a power system. In this research, symmetrical component method is used to analyze the effect of various transformer connection and generator grounding methods of single phase to ground-fault at the unit generator-transformer. Discrete Wavelet Transforms and Artificial Neural Network are applied to Ground-Fault Diagnosis Scheme at different locations at the unit generator-transformer. This faults waveform was decomposed through wavelet transform analysis into different approximations and details. A new Statistical Method and Neural Network Pattern Recognition approach, which includes statistical parameters of each type of ground-fault was used in neural network architecture for the ground-fault diagnosis. Ground-fault diagnosis scheme consists of detection and classification of ground-faults. The simulation of the unit generator-transformer was carried out using the Sim-PowerSystem Blockset of MATLAB. The statistical parameters analysis involved calculating a tendency factors including the mean, mode, median and dispersion factor including range and standard deviation values of detailed wavelet coefficients. Tendency factor and dispersion factor are used as input for Neural Network Pattern Recognition. The results of Receiver Operating Characteristic and Confusion Matrix of Neural Pattern Recognition indicated that the proposed algorithm is enough to detect and classify a ground-fault for a unit generator-transforme
展开▼
