Arc fault localization based on time-frequency characteristics of currents in photovoltaic systems

With the development of direct current (DC) distribution systems, the increasing line length makes the maintenance more difficult and the arc fault localization becomes an urgent issue. In this paper, an arc fault localization algorithm is proposed in photovoltaic systems with different loads and current levels. Firstly, based on the affine time–frequency analysis method, the proposed arc fault detection feature can accurately identify arc faults and normal states. The interference of the line impedance on arc fault detection features is studied and used to construct the arc fault localization feature. Meanwhile, due to the randomness of the arc fault, the arc fault localization feature needs to be smoothed and normalized before it can be effectively used. Then, the adaptive-network-based fuzzy inference systems (ANFIS) model is applied to predict arc fault position. The time-series generative adversarial networks method helps achieve data augmentation and improve the model accuracy. Finally, the proposed algorithm is applied on the Raspberry Pi 4b and tested online on the arc fault experimental platform. The arc fault detection accuracy reaches 100 % and the localization error is not more than 4.03 % under the condition of 0–80 m line length. The entire detection and localization time is less than 1 s, which meets the UL1699B standard.