Research On Methods Of Internal Overvoltage Identification In Distribution Network

Distribution network is directly connected to users in the whole power system,which determines it has the features with complex structure,more equipment,and large coverage area of line.The voltage level of distribution network is lower than 110kV,so its insulation level is lower than that of transmission network.Most of the accidents happened in powersystem are caused by overvoltage.But the evaluation of insulation state for distribution equipment,the improvement of power supply voltage quality and enhancement on the reliability of power supply are very dependent on the timely and accurate identification of overvoltage in distribution network.Therefore,it is of great significance to study the identification methods of overvoltage in distribution network.The experimental and simulation methods for studying overvoltage problems used by scholars both at home and abroad were introduced firstly in this paper.Based on the research of overvoltage feature expression and pattern recognition method,the signal decomposition method suitable for non-stationary signals,the basic idea of feature extraction and selection,and the classifier or neural network for feature selection were summarized.The mechanism of overvoltage in different types is analyzed.On this basis,the factors affecting the transient process of overvoltage are analyzed.By establishing 7 kinds of simulation models of overvoltage,it verified the correctness of theoretical analysis of overvoltage characteristics in distribution network,and provides a theoretical basis for the identification of overvoltage.Two methods were presented in this paper on distribution networks internal overvoltage identification.One of which based on the singular spectrum entropy,local characteristic scale decomposition and Hilbert transform(LCD-Hilbert transform)was firstly introduced.The mathematical statistics was used to obtain voltage time domain energy distribution characteristics,and the LCD-Hilbert transform combined with band-pass filtering algorithm was adopted to obtain voltage frequency domain energy distribution characteristics.In order to recognize the type of overvoltage,the characteristic threshold was determined by the Support Vector Machine(SVM).In addition,an overvoltage identification method of distribution network based on Choi-Williams Distribution(CWD)and Convolutional Neural Network(CNN)was introduced.The phase voltage spectrum was acquired by CWD,and the three-phase voltage spectrum was integrated by the image dimension reduction preprocessing method.Further more,the overvoltage identification was realized by the improved CNN with the rectangular convolution kernel.The training and testing samples were acquired by the 10kV distribution network model based on EMTP/ATP software.The test results showed that the two methods had high correct recognition rate on 7 overvoltage types in distribution network,which included.single-phase metal grounding,intermittent arc grounding,fraction-frequency resonance,fundamental-frequency resonance,high-frequency resonance,switching capacitor and switching unloaded line.Under the noise environment,the first method should combine with digital filter or the hardware filtering device in order to identify the switching overvoltage.The second method proposed in this paper did not need to construct and calculate the characteristic quantities.Compared with the first method,the second method had more advantages in robustness and adaptability,and had stronger noise resistance.The measured samples obtained from the physical experiment system of distribution network could be successfully identified by the two methods proposed in this paper.