Research On The Overvoltage Recognition Based On Polarity Discrimination And Distribution Of Transient Feature

With enlargement of power grid, rise of capacity and voltage grade, overvoltage ismore and more dangerous to transmission line and electrical equipment insulation,therefore, conducting research on overvoltage in power system is of of enormoussignificance for both safety and stability of power system. There is a variety ofovervoltage caused by different factors, and the protective measures to which varies aswell. Consequently, real-time monitoring overvoltage and distinguish fault types isnecessary for fault management and improvement of insulation coordination in powersystem.This paper researches on lightning overvoltage and internal overvoltage. Firstly,this paper studies the mechanism of lightning overvoltage, and a110kV electromagnetictransient model is set up to simulate shielding failure and back flashover. Based ontheoretical analysis and simulation result, this paper analyzes the difference betweenlightning overvoltage, and propose three types of features established by time-domainand wavelet modulus maxima analysis: polarity of insulator voltage, polarity of towercurrent and mutation polarity of tower current, establishing lightning overvoltagerecognition method based on polarity discrimination. Compared with previous lightningovervoltage recognition method, this paper introduce both insulator voltage and towercurrent as feature signal, representing the process of lightning more completeness,reflecting the essential difference between shielding failure and back flashover.Meanwhile, the feature extracted in this paper is only related to the polarity of signal,rather than detailed characteristic of lightning wave, thus free from the effect of impulsecorona and reflection of travelling wave, hence, the algorithm has strong anti-interference ability and low rates of misjudgment.Based on the mechanism and waveform characteristic of internal overvoltage, thispaper presents the distribution of transient feature of overvoltage in time domain, so asto define the feature extraction interval for each kind of overvoltage, consequentlyamplifying the difference between each other. Furthermore, this paper propose featureextraction method based on frequency domain theory, wavelet theory and singular valuedecomposition theory, extracting feature in relevant interval. Through analysis andcomparison of large amount of field data, found that an overvoltage may cause anothertype of overvoltage, which might be recorded in an overvoltage data. In consequence, the decomposition algorithm and identification method for mixed overvoltage werediscussed. This paper presents a decomposition algorithm based on wavelet modulusmaximum, defining the feature extraction interval by the distribution of waveletmodulus maximum coefficient, achieving the objective of decomposition and featureextraction.Finally, the pattern recognition method for overvoltage is discussed in the paper.Based on the distribution of transient feature of overvoltage in time domain, a codingclassification method is proposed to realize initial recognition of overvoltage. Based onleast squares support vector machine, classifiers for wavelet feature and statisticalfeature of singular value decomposition were formed. Eventually, the proposedrecognition methods were combined to construct a overvoltage recognition system, andfield data verifies the accuracy of it.