Fault Detection for Low-Voltage Residential Distribution Systems With Low-Frequency Measured Data

Increasing the number of active consumers in distribution networks necessitates transforming the current control, monitoring, and protection schemes. However, on one hand, installing high-frequency measurement devices and fast communication platforms in low-voltage (LV) distribution networks is not cost effective and scalable. On the other hand, the fault detection approaches, which can provide acceptable accuracy by relying only on low-frequency measured data (with 1-30-min sampling rates), are not developed yet. Currently, the overcurrent fault detectors work mainly based on fixed current thresholds, which makes them inefficient in a system with high-distributed-energy resources. This is due to high volatility and uncertainty in the measured profile of the current. In this article, a data-driven fault detection framework with dynamic fault current thresholds is proposed. The motivation here is to develop a framework that can locally detect and isolate faults within the LV distribution networks without requiring high-frequency sampling meters. The proposed model is based on quantile regression as a statistical method to generate the quantiles of distributions of the current measurements. Two different fault current thresholds are formulated for instantaneous and definite time fault detection schemes. The thresholds are dynamically predicted for each next time step. The proposed framework is evaluated using data from a real distribution network with 169 houses. The results suggest that the proposed model is very promising for LV residential distribution networks.