Detection of Fire-Ignition Electrical Faults for Preventing Electrical Wildfires

This paper proposes a methodology for pattern recognition and detection of electrical faults that cause fire ignition, contributing to power line situational awareness. Not necessarily all vegetation high impedance faults cause ignition. This paper aims to distinguish fire-ignition vegetation HIF from non-ignition ones using a classification method. To this end, a multi-head CNN-LSTM method is proposed. Individual heads for each feature in the time-series obtain the spatial features independently using a convolutional neural network (CNN). The features are then concatenated together and fed to a Long-Short Term Memory (LSTM) to extract the temporal features. The results show that the model can successfully detect the occurrence of a vegetation HIF and accurately classify between fire-ignition and non-ignition scenarios. The detection time of ignition HIF in the proposed approach is 1.2 seconds, which is fast enough to de-energize the line in a timely manner and prevent a wildfire from occurring.