Research on Efficient Fault Diagnosis Methods in Urban Low-Voltage Distribution Networks

Abstract

As the city's low voltage distribution network becomes more and more complex and the demand for the reliability of power supply is becoming more and more important. Traditional fault diagnosis methods often neglect the topology structure and time sequence data, which leads to bad performance when dealing with complicated faults. In this paper, an effective approach to fault diagnosis is presented by combining GCN with LSTM. Firstly, GCN is used to get the topology information of low voltage distribution network, and then the data is transferred to each node by means of graph convolution. Then, LSTM is applied to study the variation tendency of time sequence data, such as current, voltage, etc. Finally, the combination of GCN and LSTM provides a high efficiency fault diagnosis solution, which can deal with time sequence data and take into account network topology. In this paper, the performance of the proposed method is verified by simulation experiments in all kinds of failure modes. Compared with traditional signal processing and machine learning, the GCN-LSTM combination model can increase the precision of fault diagnosis by about 12%, and reduce the diagnostic time by 20%. It is proved by experiments that the method can be used to diagnose the fault efficiently and accurately in the complicated power network.