Series Arc Fault Detection And Localization In Dc Distribution Based On Master Controller

Series dc arc fault detection and localization in the modern power electronics system that comprises multiple loads is a difficult task due to the arcing noise cross-talk. The arcing noises can propagate to the adjacent loads and mistrigger the detector units. This paper proposes a master controller to compare all the Random Forest (RF) based detector units' predicted class probabilities for arc fault detection and localization. The predicted probability comes from one of the RF's attributes, where the largest probability becomes the final decision. The master controller also sends out a control signal to create synchronization between detectors. This step arranges the detectors to monitor all input currents simultaneously, and provide predicted probability correlation accurately. The master controller capability was emulated and verified using an experimental testbed of two parallel-connected constant power loads.