Spar Neuro-based routine for EHV lines

Single pole autoreclose (SPAR) has been recognized as an effective means of improvingsystem stability and reliability of electrical power systems. The most important applicationfor these systems is for critical tie lines connecting two major power sources. For stabilityconsiderations, three pole autoreclose cannot be used for these conditions. SPAR applicationsinvolve tripping only the faulted phase under single-line-to-ground fault conditions andinitiate three pole autoreclose for other types of faults. Accurate and fast phase selection ofthe faulted phase is thus of great importance in order to avoid either tripping of the incorrectphase or unnecessary three phase tripping.This thesis deals with the design of artificial neural network (ANN) based phased selector forSPAR application in extra high voltage (EHV) lines. In this approach, the spectra of the highfrequency (HF) fault generated voltage signals of the three phases only are used. Thesestransient signals are captured using a specially designed stack tuner. The motivation behindemploying the voltage signals only is to ensure complete immunity to saturation of currenttransformers (CTs) and to overcome the problems encountered in conditions with highamount of fault resistance. FFT technique is applied to a moving time-domain window of thesampled voltage signals. Training patterns for the neural network are formed from thefeatures extracted from processing the spectra of the HF voltage fault generated.The neural network employed is small in size, fast and robust. Data required for training,testing and validation of the proposed AN have been generated by ElectromagneticTransient Program (EMTP). The performance of the proposed ANN has shown promisingresults. The algorithm identifies the faulted phase in a fraction of the fundamental frequency,after the fault inception. The results of the A N output are found to be accurate under different fault conditions.