Research On Line Fault Recognition In Distribution Networks

With the construction of China's smart grid,the requirements of the users for power supply reliability are getting higher and higher.By 2020,the reliability of power supply in China should reach 99.82%.The fault of distribution network line accounts for the vast majority of users' power failures.The location and handling of faults in the distribution network is an important means of achieving the above objectives.Because most of China's distribution networks are non-effective grounding systems,so the fault component is weak,and the fault location is difficult.In recent years,the fault location technology based on the transient method has greatly improved the adaptability to the single-phase ground fault,but the fault is just located in the interval,which cannot settle difficulties to the location of the fault point.If the fault type can be identified,it can provide more detailed fault information for the fault line patrol worker,which can greatly improve the efficiency of fault investigation.Due to the complicated operating environment of the distribution network,the fault characteristics of various media such as sand and groud are various;and the presence of arc at the time of grounding further increases the complexity of the fault.In view of the above problems,this paper analyzes the fault point of the neutral resonant grounded system and the line voltage and current characteristics,and exploits the fault information.For the different types of ground faults in the 10 kV distribution network,the reliable fault waveforms data are obtained through the true test,which provides a data basis for the fault identification method research.In this paper,the fault identification method of distribution network based on phase space reconstruction and average conductance characteristics is proposed.For the arc characteristics of zero-sequence current of different medium ground faults,phase space reconstruction of zero-sequence current is used to obtain the phase plane trajectory map,and the information dimension is calculated.With the characteristics of the attractor area,the fault is classified into a reliable ground fault and an unreliable ground fault according to the feature recognition quantity;further,the average conductance of each type of fault is calculated as a fault identification criterion,and the fault is classified into wet land,dry land,wet sand,dry sand,etc.In addition,based on the ideas of firstly extraction of arc features and furtherly specific classification,the LSTM-based deep network fault arc identification method proposed in this paper uses the fault current time series as input to realize automatic extraction of arc features and obtain high accuracy classification.On this basis,the multi-layer neural network classifier is further constructed to identify various fault types,which can achieve higher classification accuracy on the test data set.The fault identification method studied in this paper can provide more abundant and effective information for the distribution network monitoring system,and provide directional guidance on fault location and fault resolution.