Research On Identification Method Of Internal Overvoltage In Power System

Internal overvoltages in power systems may cause power supply interruption,electrical equipment insulation damage and other problems.When the overvoltage occurs,it is very important to identify the type of overvoltage in time and accurately.In this paper,the mechanism,influencing factors and restricting measures of switch-on no-load line overvoltage,switch-off no-load transformer overvoltage,ferroresonance overvoltage and asymmetric grounding overvoltage are analyzed.The simulation model is built with electromagnetic transient simulation software ATPEMTP,and four simulation waveforms of internal overvoltage are obtained.Finally,the waveform characteristics are discussed.In order to solve the problem of difference of internal overvoltage among different voltage levels,the internal overvoltage data is normalized so that the internal overvoltage of different voltage levels can be compared.Due to the different overvoltage durations,it is necessary to select the unified calculation interval for the feature quantity extraction and classification identification.In this paper,the waveform characteristics and duration of the four internal overvoltages are considered.The calculation interval of the overvoltage signal is selected as the time period from 10 ms before the overvoltage occurs to 70 ms after the overvoltage occurs.Then,according to the characteristic that the three-phase voltage of switch-off no-load transformer overvoltage becomes zero in less than one period after the overvoltage occurs,the maximum effective value of the three-phase voltage in the second period after the overvoltage occurs is extracted as the characteristic quantity.After the occurrence of the asymmetric grounding overvoltage,the grounding phase voltage becomes zero and the non-grounding phase voltage rises.The minimum value of the three-phase effective voltage of one cycle is selected as the feature quantity,and a total of two time-domain feature quantities are obtained.Then,the overvoltage data are transformed by Fourier transform.According to the characteristics of the main harmonics order and the ratio of the main harmonic energy to the fundamental wave energy,the amplitude of the maximum harmonics and the maximum harmonic energy to the fundamental wave energy ratio are selected.Finally,the overvoltage data are transformed by wavelet transform,and the energy entropy of the wavelet is calculated.A total of 13 characteristic quantities such as the relative energy of wavelet and the energy entropy of the wavelet are selected.In this paper,support vector machine,BP neural network and k-nearest neighbor classifier are used to identify the types of internal overvoltage in power systems.Because the regularization parameter C and the kernel function parameter gamma of support vector machine have great influence on the recognition accuracy,the grid search method and genetic algorithm are used to optimize the two parameters respectively.Secondly,since BP neural network has strong local search ability,poor global search ability and easy to fall into local extremum,it is optimized by particle swarm optimization algorithm with strong global search ability.Thirdly,because the selection of K value of k-nearest neighbor classifier will affect the classification results and recognition accuracy,the optimal K value is selected by a specific method.The recognition results show that the genetic algorithm is effective for the two-parameter optimization of support vector machine,and has a high accuracy in the identification of internal overvoltage,but it takes a long time.The grid search method is fast and has a high accuracy in the identification of internal overvoltage.The particle swarm optimization algorithm improves the recognition accuracy and reduces the training time of BP neural network for internal overvoltage.A suitable k value enables the k-nearest neighbor classifier to have the highest accuracy for internal overvoltage identification and the shortest recognition time.