A Spatial Location Method for DC Series Arc Faults Based on RSSI and Bayesian Regularization Neural Network

Fires caused by DC series arcs are one of the primary threats to the safety of photovoltaic (PV) systems. Accurate fault location is extremely helpful in protecting PV plants, but existing research on locating DC series arc faults is limited. This paper proposes a novel DC series arc spatial location method. The proposed method estimates the distance based on the received signal strength indicator (RSSI) of electromagnetic radiation (EMR) emitted by the arc. The experimental system uses four antennas that form a tetrahedral topology to receive arc EMR signals. Some pretests are implemented before an arc occurs to obtain some known-location arc EMR signals, which are processed to be the candidate data of the arc EMR model. When the arc fault occurs, the candidate data are selected through comparison with the unknown-location signals. The arc EMR model is set up by the Bayesian regularized neural network (BRNN) algorithm. Then, the distances between the arc and antennas are estimated by the BRNN. The spatial coordinates are calculated with the antenna array topology and estimated distances. Finally, the verification results show that the algorithm has a 0.5 m average spatial location error, which is proven to be active and feasible.