Advanced in Islanding Detection and Fault Classification for Grid-Connected Distributed Generation using Deep Learning Neural Network
Nowadays, the use of renewable energy is increasing, especially distributed power generation (DG) connected to the power grid. There are several problems when DG is connected to the grid. The principal obstacle pertains to the detachment of Distributed Generation (DG) from the grid, a phenomenon wel...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Khairun University, Faculty of Engineering, Department of Electrical Engineering
2025-01-01
|
| Series: | Protek: Jurnal Ilmiah Teknik Elektro |
| Subjects: | |
| Online Access: | https://ejournal.unkhair.ac.id/index.php/protk/article/view/7573 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Nowadays, the use of renewable energy is increasing, especially distributed power generation (DG) connected to the power grid. There are several problems when DG is connected to the grid. The principal obstacle pertains to the detachment of Distributed Generation (DG) from the grid, a phenomenon well known as islanding. Islanding detection is an important task that should be completed in no more than two seconds. Earlier studies have shown several approaches to islanding detection. The use of an Artificial Neural Network (ANN) based on the learning vector quantization (LVQ) technique is proposed in this paper for fault classification and islanding detection in grid-connected distributed generators. The method consists of discrete wavelet transform (DWT), which extracts some features from the fault signal. Then, LVQ is used to classify the disturbance and detect islanding events. Power, entropy, and total harmonic distortion (THD) are used to obtain the total harmonic value. All features become inputs for LVQ, and system disturbances, lightning, and islanding disturbances are used as LVQ outputs. There are 600 datasets consisting of 200 datasets for each fault as training data. To test the LVQ training results, 120 datasets consisting of 40 datasets for each disturbance are used. The training error is made at 0.1 percent to get good testing results. The test results from 120 datasets showed that the test data achieved 99.10% accuracy. In other words, the test results are very effective because there are only 0.9% errors, and there are 2 test data that do not match the actual situation. |
|---|---|
| ISSN: | 2354-8924 2527-9572 |