DOI QR코드

DOI QR Code

합성곱 신경망을 이용한 전기 아크 신호 검출

Electrical Arc Detection using Convolutional Neural Network

  • 이상익 (한국전기안전공사 전기안전연구원) ;
  • 강석우 (한국전기안전공사 전기안전연구원) ;
  • 김태원 (한국전기안전공사 전기안전연구원) ;
  • 김만배 (강원대학교 컴퓨터정보통신공학과)
  • Lee, Sangik (Electrical Safety Research Institute & Korea Electrical Safety Corp.) ;
  • Kang, Seokwoo (Electrical Safety Research Institute & Korea Electrical Safety Corp.) ;
  • Kim, Taewon (Electrical Safety Research Institute & Korea Electrical Safety Corp.) ;
  • Kim, Manbae (Dept. of Computer & Communications Eng., Kangwon National University)
  • 투고 : 2020.03.25
  • 심사 : 2020.06.24
  • 발행 : 2020.07.30

초록

전기화재의 원인중의 하나는 직렬 아크이다. 최근까지 아크 신호를 검출하기 위해 다양한 기법들이 진행되고 있다. 시간 신호에 푸리에 변환, 웨이블릿 변환, 또는 통계적 특징 등을 활용하여 아크 검출을 하는 방법들이 소개되었지만, 변환 및 특징 추출은 부가적인 처리 시간이 요구되는 단점이 있다. 반면에 최근의 딥러닝 모델은 종단간 학습으로 특징 추출 과정없이 직접 원시 데이터를 활용한다. 따라서, 1-D 시간 신호를 직접 활용하여 아크를 검출하는 것이 좋은데, 인공신경망의 분류 성능이 저하되는 문제점이 있다. 본 논문에서는 연속 입력 1-D 신호를 2-D로 변환한 후에, 합성곱신경망으로 분류하는 방법을 제안한다. 실험 데이터에 적용한 결과 합성곱신경망의 사용이 인공신경망보다 약 8.6%의 아크 분류 성능을 향상시켰다. 또한 2-D 데이터의 부족을 보완하기 위해서 데이터증강을 이용하여, 14%의 분류 성능을 개선하였다.

The serial arc is one of factors causing electrical fires. Over past decades, various researches have been carried out to detect arc occurrences. Even though frequency analysis, wavelet, and statistical features have been used, additional steps such as transformation and feature extraction are required. On the contrary, deep learning models directly use the raw data without any feature extraction processes. Therefore, the usage of time-domain data is preferred, but the performance is not satisfactory. To solve this problem, subsequent 1-D signals are transformed into 2-D data that can feed into a convolutional neural network (CNN). Experiments validated that CNN model outperforms deep neural network (DNN) by the classification accuracy of 8.6%. In addition, data augmentation is utilized, resulting in the accuracy improvement by 14%.

키워드

참고문헌

  1. C. Wu, Y. Liu and C. Hung, "Intelligent detection of serial arc fault on low voltage power lines", J. of Marine Science and Technology, Vol. 25, No. 1, pp. 43-53, 2017.
  2. S. Ma, and L. Guan, "Arc fault recognition based on BP Neural Network", Int' Conf. Measuring Technology and Mechatronics Automation, 2011.
  3. H. Yuanhang, Y. Wang, D. Enyuan, and Z. Jiyan, "Aviation arc fault diagnosis based on weight direct determined neural network", Int. Conf. Electric Power Equipment, 2013.
  4. S. Hong, T. Kim, and S. Lee, "Study of series-arc detection algorithm", KIEE Summer Conf., 2018.
  5. N. Perera and A. Rajapakse, "Recognition of fault transients using a probabilistic neural network classifier", IEEE Trans. Power Delivery, Vol. 25, Iss. 1, 2011.
  6. Z. Chen and W. Li, "Multisensor feature fusion for bearing fault diagnosis using sparse autoencoder and deep belief network", IEEE Tran. Inst. and Measure., Vol. 66, No. 7, July 2017.
  7. H. Gu, F. Zhang, Z. Wang, Q. Ning, and S. Zhang, "Identification method for low-voltage arc fault based on the loose combination of wavelet transformation and neural network", Power Eng. and Auto. Conf., 2012.
  8. P. Muller, S. Tenbohlen, R. Maier, and M. Anheuser, "Characteristics of series and parallel low current arc faults in the time and frequency domain", Proc. of the 56th IEEE Holm Conf. Electrical Contacts, 2010.
  9. G. Yunmei, W. Li, W. Zhuoqi, and J. Binfeng, "Wavelet packet analysis applied in detection of low-voltage DC arc fault", IEEE Industrial Electronics and Applications, 2009.
  10. S. Lee, C. Choi, and M. Kim, "CNN-based people recognition for vision occupancy sensors", Journal of Broadcast Engineering, Vol. 23, No. 2, March 2018, pp. 274-282. https://doi.org/10.5909/JBE.2018.23.2.274
  11. E. Kim and W. Kim, "Face anti-spoofing based on combination of luminance and chrominance with convolutional neural networks", Journal of Broadcast Engineering, Vol. 24, No. 6, pp. 1113-1121, Nov. 2018. https://doi.org/10.5909/JBE.2019.24.6.1113
  12. T. Um, F. Pfister, D. Pichler, S. Endo, M. Lang, S. Hirche, U. Fietzek, and D. Kulic, "Data augmentation of wearable sensor data for Parkinson's disease monitoring using convolutional neural networks", 19th ACM International Conference on Multimodal Interaction (ICMI), Nov. 2017, Glasgow, UK.