Data-driven transmission line developing-fault identification based on recorded transient waveforms

Transmission lines located in complex environments are essential carriers for transmitting electricity. Most overhead transmission line faults are caused by meteorological factors. Faults caused by persistent extreme weather, such as mountain fires, icing, snowstorms, or typhoons, may develop further and thus are not suitable for reclosing attempts. In contrast, faults caused by instantaneous extreme weather, such as lightning strikes, are generally suitable for reclosing attempts. Data-driven rapid identification of whether a transmission line fault is a developing fault can assist in determining the suitability of a reclosing attempt. This study designed a preprocessing method to parse fault record files measured at a single end of the transmission line, directly extracting the fault transient voltage and current waveforms of the faulted line from the fault recorder data. A modified AlexNet was trained and tested using only extracted raw waveforms, without manual feature extraction or simulated data. The trained model can determine whether a fault caused by extreme weather is a developing fault quickly and accurately, based on the input of recorded transient waveforms.