Feed-forward Artificial Neural Network–Discrete Wavelet Transform Approach to Classify Power System Transients

Abstract

Switching transients in power systems are always a concern in studies of equipment insulation coordination. However, with widespread use of sensitive non-linear electronic devices, these transients are capable of degrading the quality of power. Utilities often switch the shunt capacitor banks to cope with sagging voltage levels, thereby generating transients that travel into the network of end users. Capacitor switching can cause over-voltage, resonance, and inadvertent tripping of adjustable speed drives and many other sensitive electronic devices. This article presents a method to distinguish between transients arising out of isolated capacitor switching, back-to-back capacitor switching, load switching, various line faults, and line switching. Discrete wavelet transform of the modal voltage signal is used to extract distinguishing features from the voltage waveform of these events. The detailed coefficients for d1 and d5 level only, obtained from discrete wavelet transform, are processed, mapped, and given to the feed-forward artificial neural network, which classifies the event accordingly. A real power system has been simulated in PSCAD/EMTDC with lines modeled using a frequency-dependent phase model, and its results are then fed to MATLAB (The MathWorks, Natick, Massachusetts, USA) for implementation of the scheme using the feed-forward artificial neural network.