| Title |
Prediction of Creepage Insulation Strength of Thermally Degraded Composites Using k-NN and ANN |
| Authors |
황인기(Inki Hwang) ; 유광열(Kwangyeol Yoo) ; 김명진(Myungchin Kim) |
| DOI |
https://doi.org/10.5370/KIEE.2024.73.12.2482 |
| Keywords |
Flashover Voltage; Degradation; Machine Learning; Artificial Neural Network; K-Nearest Neighbors |
| Abstract |
In this study, the effectiveness and performance of machine learning approaches for predicting surface flashover voltage of thermally stressed specimen were studied. In particular, k-NN (k-Nearest Neighbors) and ANN (Artificial Neural Network) technique were considered. Data from flashover experiments and features of the electrical field distribution were used to train the machine learning models. The overall process of implementing the prediction model is introduced. While both approaches were optimized through hyperparameter tuning, it was shown that the ANN approach shows slightly better prediction performance. In particular, ANN seemed to show satisfactory performance for specimen that experienced higher stress conditions. Although the k-NN approach showed some limitations compared to the ANN approach, the k-NN approach could also show potential for applications thanks to its rather simple approach and requiring less computation resources. |