Arc Fault Detection in DC Distribution Using Semi-Supervised Ensemble Machine Learning

Series arc fault detection in a dc system is a challenging task due to the randomness of arc discharge and the dynamic behavior dependence on the system current level. DC arc faults could potentially create a fire hazard if not detected and isolated quickly. This paper introduces two ways to improve the predictive performance of existing conventional machine learning (ML) algorithms as an arc fault detection method. A semi-supervised ML ensures sufficient training data when there is a vast number of unlabeled data and limited labeled data. An ensemble ML method further superimposes on a conventional ML algorithm to create a better classifier, which reduces the bias and decision variance. The goal is to evaluate the effectiveness of both methods in dc arc fault detection. Accuracy, precision, and recall scores are used as the key performance metrics. Finally, an experimental arc fault detection was conducted using a microcontroller Udoo X86 with detection latency time as a performance metric.