Detection and classification of short-circuit faults by considering the pick-up current and using the CNN-LSTM hybrid model in power distribution grids

In this paper, the authors present a technique based on a hybrid model comprising Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) architectures, referred to as the CNN-LSTM hybrid model, for the detection and classification of various types of short-circuit faults in medium-voltage power distribution grids (PDGs). Three-phase current signals were initially measured and sampled at a medium-voltage/low-voltage substation following the occurrence of a specific type of short-circuit fault. Subsequently, the convolutional layers extract spatial features from the current signals using convolutional operations, and the extracted features are then fed into the LSTM model to identify temporal patterns in the signals. Additionally, a pick-up current value, defined by the operator for the network, is considered as a threshold for the minimum RMS fault current, ensuring that both the waveform and RMS value are criteria for fault detection. For evaluation, a three-phase ۲۰ kV isolated system was implemented in MATLAB/Simulink under various fault conditions. The performance of the proposed technique was also tested in a Python environment. Based on the simulation results, several conclusions can be drawn: (i) The proposed method is a suitable option for real-time applications. (ii) This study can detect and classify faults using only current waveforms. (iii) The novel approach considers the pick-up current as a factor for fault detection.

کلیدواژه ها:

short-circuit faults ، fault detection ، fault classification ، CNN-LSTM hybrid model ، power distribution grids