Abstract
The quality of high voltage 15kV insulators could be determined by the leakage current in the transmission line, which needed to be replaced during the maintenance process. Some contaminated high-voltage insulators could have serious flash-over phenomena, which could cause interruptions in the transmission towers. In this study, the deep learning gated recurrent unit (GRU) is utilized to classify the leakage current for 15kV HDPE insulators in Taiwan. The weather parameters are also collected with the leakage current in the coastal areas which generally suffer strong wind and pollution. The Pearson matrix is deployed for selecting the most correlative features with the target leakage current. The proposed GRU is analysed and compared with the traditional recurrent neural network (RNN) by deploying the category cross-entropy (CRE), and the accuracy (AC) benchmarks. The experiment results demonstrated the significant improvements of the GRU algorithm compared with the traditional RNN model, which acquires the maximum enhancement of 25.83% CRE, 34.67% validating CRE, 23.57% AC, and 31.09% validating AC, respectively. The predicting values could contribute comprehensive information about the contamination levels of 15kV HDPE insulators, and provides meaningful data for decision-making during the maintenance operations.
Thông tin:
| Thuộc danh mục | |
| Hội thảo quốc tế | 2023 International Conference on Science, Education, and Viable Engineering |
| Tổ chức | Taiwan Association for Academic Innovation (TAAI) IEEE Electron Devices Society (IEEE EDS) University of Economics Ho Chi Minh City |
| Trang | 135 |
| ISBN | 978-604-80-7896-6 |
| Ngày tổ chức hội thảo | 12-16/04/2023 |
| Ngày xuất bản | 18/04/2023 |
| Link | ttps://drive.google.com/drive/folders/1v806uXc_xa3_sBqvAazHURcGnnmiYvEE |