Application of Machine Learning in Fault Detection And Classification in Power Transmission Lines
Electrical faults have been identified as a significant contributing factor to electrical equipment damage. Such incidents can potentially result in a range of adverse consequences, including bushfires, electrical outages, and power shortages. The detection and classification of faults facilitates t...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Pusat Penelitian dan Pengabdian Masyarakat (P3M), Politeknik Negeri Cilacap
2024-12-01
|
| Series: | Journal of Innovation Information Technology and Application |
| Subjects: | |
| Online Access: | https://ejournal.pnc.ac.id/index.php/jinita/article/view/2424 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Electrical faults have been identified as a significant contributing factor to electrical equipment damage. Such incidents can potentially result in a range of adverse consequences, including bushfires, electrical outages, and power shortages. The detection and classification of faults facilitates the delivery of superior quality of service, the preservation of the environment, the prevention of equipment damage, and the satisfaction of electricity line subscribers. In this study, we analyze the data from an electrical network comprising four generators of 11 kV, which have been modeled in Matlab. The generators are situated in pairs at either end of the transmission line. Subsequently, machine learning techniques are employed to detect faults in the transmission between lines, and machine learning models are utilized to classify the faults. Four distinct supervised machine learning classifiers are employed for comparison purposes, with the results presented in a confusion matrix. The results demonstrated that decision trees are particularly well-suited to this task, with an 88.6205% detection rate and a slightly higher accuracy than the random forest algorithm (87.9212% detection rate). The K-nearest neighbor's approach yielded a lower result (80.4196% of faults detected), while logistic regression demonstrated the lowest performance, with 34.5836% of faults detected. Six fault categories were found in the dataset: No-Fault (2365 occurrences), Line A Line B to Ground Fault (1134 occurrences), Three-Phase with Ground (1133 occurrences), Line-to-Line AB (1129 occurrences), Three-Phase (1096 occurrences) and finally Line-to-Line with Ground BC (1004 occurrences). |
|---|---|
| ISSN: | 2716-0858 2715-9248 |