Self-Attention Network for Partial-Discharge Diagnosis in Gas-Insulated Switchgear

Abstract

Detecting, measuring, and classifying partial discharges (PDs) are important tasks for assessing the condition of insulation systems used in different electrical equipment. Owing to the implementation of the phase-resolved PD (PRPD) as a sequence input, an existing method that processes sequential data, e.g., the recurrent neural network, using a long short-term memory (LSTM) has been applied for fault classification. However, the model performance is not further improved because of the lack of supporting parallel computation and the inability to recognize the relevance of all inputs. To overcome these two drawbacks, we propose a novel deep-learning model in this study based on a self-attention mechanism to classify the PD patterns in a gas-insulated switchgear (GIS). The proposed model uses a self-attention block that offers the advantages of simultaneous computation and selective focusing on parts of the PRPD signals and a classification block to finally classify faults in the GIS. Moreover, the combination of LSTM and self-attention is considered for comparison purposes. The experimental results show that the proposed method achieves performance superiority compared with the previous neural networks, whereas the model complexity is significantly reduced.

One-Shot Learning for Partial Discharge Diagnosis Using Ultra-High-Frequency Sensor in Gas-Insulated Switchgear

“…In this section, we present the PRPD and noise measurements that were obtained using a PD monitoring system for the GIS [ 27 , 28 ]. Figure 1 shows a block diagram of the PD monitoring system, consisting of a GIS, external UHF sensor, amplifier, a peak detector, and a data acquisition system (DAS).…”

“…Deep neural networks, which combine feature extraction and classification, have achieved promising results in several application areas, such as computer vision, natural language processing, speech recognition, and text classification [ 25 ]. Deep learning models, such as convolutional neural networks (CNNs) [ 26 ], recurrent neural networks (RNNs) [ 27 ], and self-attention [ 28 ], have been employed and state-of-the-art results have been obtained, in order to improve the performance for fault diagnosis while using a PRPD in a GIS. [ 26 ] proposed a CNN to learn the local response from the temporal or spatial signals of a PRPD.…”

“…[ 27 ] proposed an RNN model with a long short-term memory to process sequential PRPD data. To overcome parallel computation and enable the capture of the interactions among PRPDs, [ 28 ] used a self-attention neural network to assess the possibility of focusing on important information and simultaneous computation from PRPD input. However, most of the existing deep-learning-based fault diagnosis methods require a large dataset for training and validation [ 26 , 27 , 28 ].…”

“…To overcome parallel computation and enable the capture of the interactions among PRPDs, [ 28 ] used a self-attention neural network to assess the possibility of focusing on important information and simultaneous computation from PRPD input. However, most of the existing deep-learning-based fault diagnosis methods require a large dataset for training and validation [ 26 , 27 , 28 ]. However, it is difficult to obtain a large dataset for fault diagnosis in the real world [ 27 , 28 ].…”

“…However, most of the existing deep-learning-based fault diagnosis methods require a large dataset for training and validation [ 26 , 27 , 28 ]. However, it is difficult to obtain a large dataset for fault diagnosis in the real world [ 27 , 28 ].…”