Power System Transients Characterization and Classification Using Wavelets and Neural Networks

s This paper presents a framework for transients classification using wavelet transform and two specific artificial neural networks, propabilistic neural networks (PNN) and resource allocating neural networks (RAN). One significant feature of wavelet for transients characterization is its timescale two dimensional but compact representation. One distinguishing feature of PNN and RAN is the ability to adjust their architecture automatically to adapt to new environment quickly and accurately, which makes them the promising candidates for transients classification in power system. Some experimental results have indicated the suitability of this framework for transients classification. I. INTRODUCTION Electromagnetic transients in power systems result from a variety of disturbances on transmission lines, such as switching , lightning strikes, faults, as well as from other intended or unintended events. Such transients are extremely important, for it is at such times that the power system components are subjected to the greatest stresses from excessive currents or overvoltages[KiSh83][MoKi96]. In recent years, an increasing demand has been observed for power system monitoring and transient recording (including faults) combined with analysis, classification, and reporting. One important component of such system is the digital transient recorder, which has been available in some well-equipped substations to record various transients and provide assistant information about the transients for system operators to analyze them and take actions whenever a fault occurs. However, present transient recorders lack classification capability to distinguish " interesting " events from " trivial " events. The desired intelligent transient recorder should have the ability to distinguish and classify different kinds of tran-sients, provide a means to filter out meaningful information needed to assess system performance, classify faults and other disturbances to help identify stability, fault or other power quality issues, and even more importantly to provide a means to look automatically for early anomalies that could cause catastrophic failure. Transients are signals with a finite life, that is, a transient dies to zero in a finite time. Frequency based analysis has been common since Fourier's time; however, frequency anal