Development of an Intelligent Detection Method of DC Series Arc Fault in Photovoltaic System Using Multilayer Perceptron and Bi-Directional Long Short-Term Memory

A DC series arc fault is one of the significant sources of electrical wiring fires in residential buildings. The production of extremely high temperatures may lead to the ignition of nearby combustible materials. The applications of arc fault diagnosis based machine learning are a global interest due to the immense challenge to create an accurate and efficient detection method. In this paper, a detection and classification method using a multilayer perceptron incorporated with Bi-Directional Long shortterm Memory (MLP-BiLSTM) is proposed. In order to achieve this goal, nine series arc fault models are used in conjunction with data from real-world observations for simulation purposes using Power System Computer Aided Design (PSCAD) software. The simulation and experimental results confirm that the accuracy of the proposed detection and classification method reaches 99%, which results in that the methodology is believed to be accurate for DC series arc fault detection and classification in the PV system with relatively high accuracy.