Power transmission system’s fault location, detection, and classification: Pay close attention to transmission nodes

• Published in: International journal of electrical power & energy systems 2024-02, Vol.156, p.109771, Article 109771
• Main Authors: Ukwuoma, Chiagoziem C., Cai, Dongsheng, Bamisile, Olusola, Chukwuebuka, Ejiyi J., Favour, Ekong, Emmanuel, Gyarteng S.A., Caroline, Acen, Abdi, Sabirin F.
• Format: Article
• Language: English
• Subjects: Attention mechanism; Fault diagnosis; Fault location; Graph neural network; Multi-linear perceptron; Transmission system
• ISSN: 0142-0615
• DOI: 10.1016/j.ijepes.2023.109771

•A new fault diagnostic is proposed for fault localization, detection, and classification in power transmission systems.•Adds spatial characteristics to the training procedure, allowing the model to derive a richer topological understanding from the data and show more resistance to anomalous data and varying circumstances.•An MSA Network is designed to extract fault characteristics, enhancing the model's data-analyzing abilities.•An MLP Network that uses non-linear mapping abilities, characteristic representation competencies, and abstraction power is designed. For transmission systems to operate safely and reliably, fault identification and classification are essential. However, power network physical architecture and data information cannot be fully utilized by conventional intelligent approaches. This study,therefore,presents a fault localization, detection, and classification model for transmission systems that concentrate on the key distribution nodes. The model makes use of a deep graph neural network with multi-scale attention and multi-linear perceptron block which accounts for the power network's structural composition during learning. The model's capacity to manage unusual data input and unidentified application situations is improved by the inclusion of multi-scale attention. Furthermore, it enables the model to precisely pinpoint fault areas by identifying patterns and connections among system parts, concentrating on specific areas or nodes. In addition, a multi-linear perceptron block is designed to enhance the capturing of amplitude information and increase comprehension. The efficiency and generalizability of the proposed model are improved by the implementation of a multi-task training approach for locating faults and their type. With the use of two IEEE 13-Bus systems and the PSS/E 23-Bus system, the proposed fault diagnosis model is tested. Examining various setups for fault analysis allows for a more thorough evaluation of the model's ability to generalize and disturbance resilience. Experimental findings show that the proposed model outperforms existing cutting-edge techniques in terms of efficacy with a balanced accuracy of 0.8204 for classification, 0.556 for localization, and a Macro MAE of 38.780 for detection.