A Communication-free Protective Relay for a Closed-Loop Distribution System Using Supervised Learning
- Abstract
- This thesis presents a new protective relay trained with machine-learning for a closed-loop distribution system. It serves as primary protection against electrical faults and classifies the fault types. Recently, as the demand for reliable high-quality power has increased, normally closed-loop distribution networks become preferred to radial networks; the protection system for a closed-loop distribution system is necessary. The protection system of a closed-loop distribution system in South Korea is incorporated into distribution automation system (DAS); it features a primary protection algorithm relying on peer-to-peer communication between protective relays. Therefore, malicious cyberattacks and communication errors can result in failures of primary protection and cause operators of the DAS to fail to take urgent measures against electrical faults. Moreover, the conventional protective relay for a closed-loop distribution system can not guarantee isolation of a faulty section in case of failures in communications as well as fault direction determination. The proposed protective relay can replace the conventional algorithm (that depends on communication and fault direction information for protection) with supervised learning (long short-term memory [LSTM]). To apply machine-learning, long short-term memory (LSTM) was used, and 12 types of time-series electrical data measured and calculated by each relay of a test power system were employed as the input to neural networks. After hyperparameters for LSTM networks were adjusted to enhance relay performance, all relays were trained with 6,000 cases and tested with 3,000 cases, respectively. Simulations showed that the proposed protective relay afforded better performance than relays having the conventional protection algorithm.
- Author(s)
- Sung-Ryeol Han
- Issued Date
- 2022
- Type
- Thesis
- URI
- https://scholar.gist.ac.kr/handle/local/18824
- 공개 및 라이선스
-
- 파일 목록
-
Items in Repository are protected by copyright, with all rights reserved, unless otherwise indicated.