Answers are grounded in this paper and cite their sources.
This method identifies and isolates faults in distribution networks, suggesting improved protection efficiency and accuracy.
As distributed power sources continue to be connected to the distribution grid, it brings serious challenges to the distribution network protection configuration. To address the challenges of balancing model efficiency with accuracy and improving the discrimination accuracy between internal and external faults, this paper proposes a dual‐channel deep network–based pilot protection method for distribution networks to achieve rapid and precise fault isolation. First, this paper analyses the transient characteristics of distribution network faults in terms of distributed power supply faults and synchronous generator faults and clarifies the differences of time‐domain transient current waveforms. Secondly, considering the obvious before‐and‐after dependence of the post‐fault current data in the time dimension, a recurrent neural network long short‐term memory (LSTM) with a gated structure is constructed. To utilise the post‐fault current data in each time range, a bidirectional LSTM (Bi‐LSTM) method is proposed to predict the current signal throughout the fault. Based on this, Kolmogorov–Arnold network (KAN) is added and the parameter dimensions are reduced to enhance the training efficiency of the longitudinal protection model. Finally, a parallel dual‐channel deep network architecture based on the KAN network and Bi‐LSTM network is designed to achieve the accurate identification and ensure the normal operation of existing relay protection devices. Results of the experiments indicated that the proposed method for distribution networks has good generalisation ability and can quickly determine the fault type and area, compared with the traditional method of protection devices, and the fault location is accurately determined at the moment of the fault and quickly isolated. The fault identification accuracy is as high as 98.7%, which validates the rationality and strength of the proposed method.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: