Deep learning for partial-discharge detection in power systems
Partial discharge is an external manifestation of the aging of the insulation of electrical equipment. The correct detection of the partial discharge signal can help determine the insulation status of electrical equipment. In the early stage of the failure, the discharge signal caused by the partial...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis-Master by Coursework |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/155097 |
| _version_ | 1826112899846766592 |
|---|---|
| author | Song, Fang |
| author2 | Wang Lipo |
| author_facet | Wang Lipo Song, Fang |
| author_sort | Song, Fang |
| collection | NTU |
| description | Partial discharge is an external manifestation of the aging of the insulation of electrical equipment. The correct detection of the partial discharge signal can help determine the insulation status of electrical equipment. In the early stage of the failure, the discharge signal caused by the partial defect is very weak. Traditional preventive tests are difficult to detect abnormal signals. Based on this, this paper combines the SVDD algorithm and the LSTM network to build the partial discharge signal recognition model. Its research work mainly includes the following two aspects:
1. Establish a sequence classification prediction model based on LSTM. This paper builds an LSTM network model to classify signals. On this basis, we try and explore feature engineering. In order to explore the performance of LSTM, we tried single-layer LSTM and double-layer LSTM networks, and analyzed the results.
2. Establish a partial discharge signal classification model based on SVDD. This article derives the SVDD algorithm in detail and introduces wavelet denoising, and tries to combine the two to improve the performance of the model.
In order to better compare whether the effect is improved, establishing the traditional PRPD method as a benchmark to test the effect of the two models. It is found that the traditional method requires manual judgment, while the first two methods can be automatically implemented after deployment and have good results. At the same time, the neural network needs a well-designed network and appropriate parameters, and the SVDD model needs better feature engineering. |
| first_indexed | 2024-10-01T03:14:42Z |
| format | Thesis-Master by Coursework |
| id | ntu-10356/155097 |
| institution | Nanyang Technological University |
| language | English |
| last_indexed | 2024-10-01T03:14:42Z |
| publishDate | 2022 |
| publisher | Nanyang Technological University |
| record_format | dspace |
| spelling | ntu-10356/1550972023-07-04T16:36:06Z Deep learning for partial-discharge detection in power systems Song, Fang Wang Lipo School of Electrical and Electronic Engineering ELPWang@ntu.edu.sg Engineering::Electrical and electronic engineering Partial discharge is an external manifestation of the aging of the insulation of electrical equipment. The correct detection of the partial discharge signal can help determine the insulation status of electrical equipment. In the early stage of the failure, the discharge signal caused by the partial defect is very weak. Traditional preventive tests are difficult to detect abnormal signals. Based on this, this paper combines the SVDD algorithm and the LSTM network to build the partial discharge signal recognition model. Its research work mainly includes the following two aspects: 1. Establish a sequence classification prediction model based on LSTM. This paper builds an LSTM network model to classify signals. On this basis, we try and explore feature engineering. In order to explore the performance of LSTM, we tried single-layer LSTM and double-layer LSTM networks, and analyzed the results. 2. Establish a partial discharge signal classification model based on SVDD. This article derives the SVDD algorithm in detail and introduces wavelet denoising, and tries to combine the two to improve the performance of the model. In order to better compare whether the effect is improved, establishing the traditional PRPD method as a benchmark to test the effect of the two models. It is found that the traditional method requires manual judgment, while the first two methods can be automatically implemented after deployment and have good results. At the same time, the neural network needs a well-designed network and appropriate parameters, and the SVDD model needs better feature engineering. Master of Science (Communications Engineering) 2022-02-07T07:02:24Z 2022-02-07T07:02:24Z 2021 Thesis-Master by Coursework Song, F. (2021). Deep learning for partial-discharge detection in power systems. Master's thesis, Nanyang Technological University, Singapore. https://hdl.handle.net/10356/155097 https://hdl.handle.net/10356/155097 en application/pdf Nanyang Technological University |
| spellingShingle | Engineering::Electrical and electronic engineering Song, Fang Deep learning for partial-discharge detection in power systems |
| title | Deep learning for partial-discharge detection in power systems |
| title_full | Deep learning for partial-discharge detection in power systems |
| title_fullStr | Deep learning for partial-discharge detection in power systems |
| title_full_unstemmed | Deep learning for partial-discharge detection in power systems |
| title_short | Deep learning for partial-discharge detection in power systems |
| title_sort | deep learning for partial discharge detection in power systems |
| topic | Engineering::Electrical and electronic engineering |
| url | https://hdl.handle.net/10356/155097 |
| work_keys_str_mv | AT songfang deeplearningforpartialdischargedetectioninpowersystems |