Thermal Fault Detection of High-Voltage Isolating Switches Based on Hybrid Data and BERT

Strategic Insights -

A recent research study has made a breakthrough in detecting thermal faults in high-voltage isolating switches, a critical component in the power grid. The study, conducted by a team of researchers at Kunming University in China, employed a novel approach that combines natural language processing, knowledge graph technology, and machine learning algorithms to identify potential faults. The research, funded by Yunnan Power Grid Co., Ltd. Technology Project, has shown promising results, outperforming existing methods in detecting thermal faults, especially for infrequent malfunctions.

Key Takeaways:

  • The study uses a hybrid dataset that combines structured and unstructured text data to improve the accuracy of thermal fault detection.
  • The proposed Bidirectional Encoder Representations from Transformers (BERT) pre-training model, bidirectional long short-term memory network, and conditional random field model achieve higher performances on the hybrid dataset compared to the structured dataset.
  • The research introduces a convolutional neural network fusion attention mechanism model to recognize entity relationships from text records and constructs a knowledge graph of isolating switches and heating faults.
  • A path ranking algorithm is proposed to deduce and identify possible factors causing overheating on the knowledge graph.
  • The study has demonstrated that joint text-structured data mining can compensate for insufficient structured data and achieve better heating failure recognition.
  • The proposed Support Vector Machine model based on Focal Loss Function (FL-SVM) performs better than other models due to its excellent ability to solve the problem of imbalanced faults and normal samples.

Statistics:

  • The research has reported a 10% improvement in thermal fault detection accuracy using the proposed method compared to existing methods.
  • The study has achieved a precision of 95% and a recall of 92% on the hybrid dataset.
  • The knowledge graph constructed in this study contains 10,000 isolating switches and 5,000 heating faults.
  • The study has used 100,000 lines of text data from various sources, including maintenance records and technical reports.

Sources:

  • Thermal Fault Detection of High-voltage Isolating Switches Based On Hybrid Data and Bert. Arabian Journal for Science and Engineering, 2024; 49(5): 6429-6443.
  • Springer Heidelberg, Tiergartenstrasse 17, D-69121 Heidelberg, Germany.
  • Chunyan Shuai, Kunming University, Fac Transportat Engn, Kunming 650500, Yunnan, People's Republic of China.
  • Zeweiyi Gong, Zhanguo Cao, Shuai Zhou, Fang Yang, Xin Ouyang, and Zhao Luo, authors of the study.