Applications of wavelets and neural networks for classification of power system dynamics events
This paper investigates the possibility of classifying power system dynamics events using discrete wavelet transform (DWT) and a neural network (NN) by analyzing one variable at a single network bus. Following a disturbance in the power system, it will propagate through the system in the form of low-frequency electromechanical oscillations (LFEOs) in a frequency range of up to 5 Hz. DWT allows the identification of components of the LFEO, their frequencies, and magnitudes. After determining the energy components' share of the analyzed signal using DWT and Parseval's theorem, the input data for the classification process using a NN are obtained. A total of 5 classes of disturbances, 3 different wavelet functions, and 2 different variables are tested. Simulation results show that the proposed approach can classify different power disturbance types efficiently, regardless of the choice of variable or wavelet function.