Single-ended Fault Location Method Based on Traveling Wave Panoramic Fault Characteristics Deep Mining
The existing fault location methods only use the local fault characteristic of traveling wavefront,which results in fault location failure at weak faults(zero crossing fault,high impedance fault)and close-in faults.Therefore,a single-ended fault location method based on traveling wave panoramic fault characteristics deep mining is proposed in this paper.Firstly,based on the time-frequency domain traveling wave full waveform(TWFW),it is proved theoretically and simulated that traveling wave arrival sequence in time domain can reflect different fault sections,and the frequency distribution of traveling wavefronts can reflect the fault location.The mapping mechanism between TWFW and fault distance is qualitatively analyzed,and the uniqueness theory of TWFW is demonstrated.Then,taking the TWFW as the input of convolution neural network(CNN),the CNN referred from the lightweight LeNet-5 is built.The 3×3 small size convolution kernel is used to mining the panoramic fault characteristics of TWFW.The mapping relationship between the panoramic fault characteristics of TWFW and fault distance is established,so as to realize accurate fault location.Finally,the Grad-CAM visualization method is utilized to show the fault sensitive feature of TWFW mined in each convolution channel of CNN.It strongly demonstrates the internal reason for the robustness of the proposed method.The simulation results show that the proposed method has high fault location accuracy,especially for weak fault and close-in fault.The average absolute error of fault location is 99.855 m.
国家科技期刊平台
NSTL
万方数据
