Protection Scheme based on Artificial Neural Network for Fault Detection and Classification in Low Voltage PV-Based DC Microgrid

Abstract

With the expansion of the DC distribution market, protection, and operational concerns for Direct Current (DC) Microgrids have increased. Different systems have been investigated for detecting, finding, and isolating defects utilising a variety of protective mechanisms. It might be difficult to locate high-resistance faults and shorted DC faults on low-voltage DC (LVDC) microgrids. Therefore, in this study, a Field Transform Technique like Short-Time Fourier Transform (STFT) is proposed for detecting the Fault Current (FC). This method detects the faults Pole-ground (PG), pole-pole (PP), and Arc fault are the major fault types in the DC network with PG fault as the most common and less severe. One of the difficulties the DC system faces in the incidence of a malfunction is the protection of essential converters. During this fault, the diodes, being the most vulnerable component of the system, may encounter a substantial surge in current, which can potentially cause damage if the current surpasses double their specified capacity for withstanding. After the Fault detection (FD), a Taguchi-based ANN is presented to classify the detected faults. This method effectively classifies PV-based faults. Then, to safeguard the FC, the Improved Self-Adaptive Solid State Circuit Breaker (I-SSCB) is introduced. It safeguards the FC in the low-voltage PV-based DC microgrid (DCMG) and restricts FC in the DCMG. The suggested approach is evaluated using the Matlab software and the proposed method produces 400A current and 100 KW power during the PV temperature of 25°C. The output current of the ANN is then 1A for a duration of 0.3 to 0.4 seconds. The fault voltage and FC produced in this proposed work are 1900V and 1950A. Therefore, the proposed work's current and voltage values are 21 KV and 0.35 I. Therefore, the proposed method produces more power and limits the FC in the LV-DCMG. In future studies, the improved or modified neural network or machine learning (ML)-based techniques can be utilized which may improve the protection scheme of the work.