CNN based Transformer Model for Fault Detection in Power System Networks

• Published in: IEEE transactions on instrumentation and measurement 2023-01, Vol.72, p.1-1
• Main Authors: Thomas, Jibin B., Chaudhari, Saurabh G., Shihabudheen, K. V., Verma, Nishchal K.
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Coders; Correlation coefficients; Data models; Deep learning; Fault detection; Faults; Feature extraction; High impedance; High impedance fault; Impedance; Machine learning; Model accuracy; Power lines; Transformer model; Transformers
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2023.3238059

Fault detection and localization in electrical power lines has long been a crucial challenge for electrical engineers as it allows the detected fault to be isolated and recovered promptly. These faults, if neglected, can rupture the normal operation of the network and drastically damage the power lines and the equipments attached to it. Wastage of power and money due to these faults can be harmful to the economy of an industry or even a country. Therefore, efficient fault detection mechanisms have become crucial for the wellbeing of this power-hungry world. This research presents an end-to-end Deep Learning strategy to detect and localize symmetrical and unsymmetrical faults as well as high-impedance faults in a distribution system. This research proposes a novel Deep CNN Transformer model to automatically detect the type and phase of the fault as well as the location of the fault. The proposed model utilizes 1-D Deep CNNs for feature extraction and Transformer Encoder for Sequence Learning. The Transformer encoder utilizes attention mechanism to integrate the sequence embeddings and focus on significant time steps to learn long-term dependence in order to extract context of the temporal current data. The different faults were simulated in MATLAB Simulink using IEEE 14-bus system. The proposed models were found to produce better performance on the test database when evaluated using F1-Score, Matthews Correlation Coefficient (MCC) and Accuracy. The models also produced better predictions on High-Impedance faults compared to conventional fault detection techniques.