A CNN-Based Fault Section Location Method in Distribution Network Using Distribution-Level PMU Data

To improve the reliability of power supply, medium and low voltage distribution network always uses small current neural grounding in China. When the single phase to ground fault occurs, the fault feature is weak and fault section is hard to locate by traditional method. The development of distribution level phasor measurement units (D-PMUs) makes the visual and precise synchronous data available. In this paper, a new fault section location method based on Convolutional Neural Network (CNN) is proposed. The data cross-section is the input characteristic matrix of CNN model corresponding to the snapshot of the synchronous measurements of multi D-PMUs at a specific time, including current and voltage phasor. And the fault section and confidence of results would be given by the trained three-layer CNN model. This method only requires the snapshot data of single moment in the steady fault stage, so it is not affected by the fault initial phase angle and is robust to the noise and data loss or corruption. The accuracy, anti-interference capacity, and effectiveness of proposed method had been tested on a 20-node distribution network with distributed generation in PSCAD.