The Study On Over-voltage Layered Pattern Identification And Application In Power System

With the scale of power grid enlarging, conveying capacity and voltage levelsrising, over-voltage is more and more seriously dangerous to transmission and electricalequipment insulation in power system, especially to UHV and EHV grids, so theresearch on over-voltage in power system is very important for the safe and reliableoperation of electrical network. There is a variety of over-voltage with different causes,so constructing an perfect over-voltage intelligent on-line monitoring system，real-timemonitoring various over-voltage signals，fast and accurately discriming the fault types,is very necessary for trouble shooting and disaster prevention in power system.However, the existing over-voltage on-line monitoring devices are still lack of theability of intelligent analysis and identification, the research on the methods ofover-voltage signals feature extracting and pattern recognition is quite significative forrestraining over-voltage, improving insulation coordination and promoting thedevelopment of smart grid.According to the traditional classification of over-voltage, firstly this paper studiesthe cause of all kinds of over-voltage, makes use of ATP-EMTP software to simulateshielding failure and back flashover over-voltage and analyses the waveformcharacteristic of common over-voltage in engineering practice. Considering the differentover-voltage of subordination relation and waveform characteristic, a new over-voltagelayered identification structure is proposed. Unlike the traditional identification system,this paper uses the top-down layer idea and establishes classifier in each layer, whichdirectly focuses on feature extraction, feature analysis and pattern recognition toimprove the real-time of over-voltage identification. Meanwhile, each classifier isindependent from others, which of program is easy to be modification and expanded.Based on over-voltage waveform characteristic, this paper extracts over-voltagefeature by time domain analysis, frequency domain analysis, wavelet analysis andsingular value decomposition (SVD) and so on. Fourier transform (FT) and wavelettransform (WT) are introduced, and frequency domain and wavelet time-frequencyfeature are extracted in this paper. In consideration of the dispersibility of switching andlightning over-voltage，a new method of multi-scale time-frequency matrix singularvalue decomposition over-voltage feature extraction is proposed by combining wavelettransform and singular value decomposition, which is used to lower the effect of dispersibility. And on that basis, this paper selects the feature of each layer, and adoptsprincipal component analysis to deal with the feature of every classifier by consideringthe correlation of feature，which lowers the number of feature and eliminates the effectof correlation.Finally this paper introduces the least square support vector machine (LS-SVM)and grid search optimization algorithm (GS), presents an over-voltage layeredidentification system based on grid search optimizing the least square support vectormachine and designs the man-machine interface for this system. Based on structure riskminimization principal, the least square SVM is more suitable for the classificationproblem of small samples, which transforms nonlinear problem to linear problem byintroducing kernel function and makes use of least square method to resolute this linearproblem, so this algorithm convergence rate quick and not existing local optimalsolution. Because of parameters selection of LS-SVM lack of theoretical guidance, thispaper employs GS algorithm to optimize the parameters of LS-SVM, which iscompared with GA and PSO optimization algorithm. The field data testing indicates thatthe method of feature extraction, feature analysis and pattern identification presented inthis paper can effectively recognize over-voltage signals. On above of all, this paperdesigns an over-voltage layered identification GUI system based on the graphical userinterface of MATLAB, combining with over-voltage on-line monitoring software andquite easy for engineers and technicians to analyze and find the fault reason, which hasgreat engineering application value.