Research On Recognition Methods For Grounding Arc And Ferroresonance Over-voltage In Distribution Power System

Overvoltage in power system has a serious impact on the safe and stable operation of electrical equipments, however, it can even cause serious accidents. Arc grounding overvoltage and ferromagnetic resonance overvoltage are two types that occur frequently in neutral non-grounding power distribution network. The similarities shared by the two types of overvoltage in time of duration, features of dispatching terminal and outcomes of accidents, it is easy for operators to make wrong judgments and thus fail to take effective measures to prevent and suppress the occurrence of overvoltage, therefore, it is necessary to make specific identification study on the two types of overvoltage. This thesis, in view of arc grounding overvoltage and ferromagnetic resonance overvoltage, based on the analysis of mechanism and research on the simulation of voltage signal characteristics, makes a classified identification study on these two type overvoltage through the application of GA-optimized BP neural network.First of all, the thesis makes an analysis on the generation mechanism of arc grounding overvoltage, and builds a simulation model of arc grounding overvoltage in lOkV neutral non-grounding power distribution network. Through the simulation, the author of the thesis acquires the waveforms of arc grounding overvoltage. The author analyzes the amplitude and spectrum of the voltagesignals and makes a simulation study on the suppressing measures against arc grounding overvoltage. The result shows that arc grounding overvoltage has high transient vibration amplitude, obvious impact performance, severe wave distortion, complicated frequency spectrum composition and large proportion of high-frequency component. It can be suppressed by means of neutral point arc suppression coil grounding and low resistance grounding.Next, the thesis makes an analysis on the generation mechanism of ferromagnetic resonance overvoltage and builds a simulation model of ferromagnetic resonance overvoltage in 10kV neutral non-grounding power distribution network by use of Matlab/Simulink. The simulation provides the waveforms of fundamental frequency, subharmonic, harmonic ferromagnetic resonance voltage and no resonance voltage. It also provides the waveform of the primary current going through potential transformer (PT). The author of the thesis analyzed the amplitude and frequency spectrum of the signals of voltage and current, and conducts simulation study on the suppressing measures against ferromagnetic resonance overvoltage. It shows that ferromagnetic resonance overvoltage consists of relatively single spectral component, but it also shows some distortion. Ferromagnetic resonance overvoltage can be effectively suppressed by the neutral point grounding of PT primary side through single-phase PT and nonlinear resistance, PT open triangle injected into damping resistance and system neutral grounding through low resistance.Lastly, the thesis proposes an identification method based on GA-oPTimized BP neural network. Based on the features of overvoltage, the author finds 14 elements, including the effective value, kurtosis, wavelet energy entropy of three-phase voltage and zero sequence voltage as well as the distribution features of wavelet energy of zero sequence. Considering the traditional BP neural network based on gradient descent learning algorithm has the defects of low convergence rate and easiness to fall into local minimum value, the author uses genetic algorithm to optimize the initial weight value and threshold value of BP neural network and use least square algorithm to optimize the learning algorithm. Then he detects the features of arc grounding overvoltage and ferromagnetic resonance overvoltage and conducts identification test through the traditional and optimized BP neural network. The result shows that the optimized BP neural network can accurately identify arc grounding overvoltage and ferromagnetic resonance overvoltage.