Fault Cause Identification For Transmission Lines Based On Recorded Data And Deep Learning

The power grid is an important infrastructure related to national economy and people's livelihood.The safe and stable operation of power system plays a vital role in social and economic development and people's normal life.With the improvement of electrical equipment manufacturing technology and field operation and maintenance level,the main causes of power system failure have been changed into natural factors such as lightning strike,mountain fire,bird damage and man-made factors such as construction collision line.Transmission lines,as an important part of power system,cover a wide area and are exposed to the field for a long time.Transmission line tripping will not only cause an impact on power equipment,but also cause a large-scale power failure in severe cases,threatening the safety of life and property.When the fault occurs,due to the quick action of the relay protection device,the fault traces may not be very obvious,which poses a challenge to the operation and maintenance personnel to quickly find the fault cause.It is not only possible to quickly restore the power supply and reduce the loss of power failure,but also conducive to taking targeted protection measures in key areas,fundamentally eliminating security risks and improving the operation and maintenance level of the power grid.At present,there are few researches on the causes of transmission line faults or the identification objects are relatively simple,so there is a lack of systematic identification methods.This paper mainly studies on six single-phase ground faults of transmission lines caused by lightning strike,bird damage,pollution flash,mountain fire,tree flash and construction collision line.Based on the analysis of fault mechanism caused by different causes,combined with fault information provided by recorded wave data,the complex mapping relationship between fault characteristics and fault causes was found,and the most concise and effective identification features were selected.The environmental characteristics provided in the recording files include season and time period information,and the distribution law can be obtained by mathematical statistics.Electrical characteristics include waveform changes of current during the fault,DC content,change of transition resistance,time and frequency distribution of the third harmonic by processing the waveform data.Because the fault label of recorded data is incomplete,the deep learning algorithm depth belief network which can be used for unsupervised learning is adopted as the classification model,and the recorded data is pretreated with synthetic minority oversampling technology(SMOTE)to reduce the unbalance of samples.The electrical characteristics selected in the previous paper are directly taken as the input,and the six types of fault causes are output.The deep belief network extracts the input data layer by layer through the multi-layer network to build the correlation model between the input and output.The input does not need to go through intermediate processing steps such as weighting,which reduces the systematic error of classification model.The output is presented in the form of probability.Combined with the statistical probability of the aforementioned environmental features,DS evidence theory is used to fuse the two results,so as to achieve the effect of both environmental features and electrical features,and improve the fault tolerance of the identification method.This paper takes a large number of different types of recorded wave data as an example and uses Matlab to carry out simulation verification for the comprehensive identification method of fault causes proposed in this paper.The analysis results show that method of fault cause identification for transmission lines based on deep learning and recorded wave data adopted in this paper has good identification ability,can accurately and effectively establish the mapping relationship between recorded wave information and fault cause,and can provide decision support for the quick search of fault cause,and meet the engineering needs.