Logistic Regression Based Arc Fault Detection in Photovoltaic Systems Under Different Conditions

doi:10.1007/s12204-019-2095-1

Abstract

Abstract: This paper investigates direct current (DC) arc fault detection in photovoltaic system. In order to avoid the risk of fire ignition caused by the arc fault in the photovoltaic power supply, it is urgent to detect the DC arc fault in the photovoltaic system. Once an arc fault is detected, the power supply should be cut off immediately. A lot of field experiments are carried out to obtain the data of arc fault current of the photovoltaic system under different current conditions. Cable length, arc gap, and the effects of different sensors are tested. These three conditions are the most significant features of this paper. Four characteristic variables from both the time domain and the frequency domain are extracted to identify the arc fault. Then the logistic regression method in the field of artificial intelligence and machine learning is originally used to analyze the experimental results of arc fault in the photovoltaic system. The function between the probability of the arc fault and the change of the characteristic variables is obtained. After validating 80 groups of experimental data under different conditions, the accuracy rate of the arc fault detection by this algorithm is proved to reach 100%.

Key words: photovoltaic| arc fault| fast Fourier transform| logistic regression

摘要： This paper investigates direct current (DC) arc fault detection in photovoltaic system. In order to avoid the risk of fire ignition caused by the arc fault in the photovoltaic power supply, it is urgent to detect the DC arc fault in the photovoltaic system. Once an arc fault is detected, the power supply should be cut off immediately. A lot of field experiments are carried out to obtain the data of arc fault current of the photovoltaic system under different current conditions. Cable length, arc gap, and the effects of different sensors are tested. These three conditions are the most significant features of this paper. Four characteristic variables from both the time domain and the frequency domain are extracted to identify the arc fault. Then the logistic regression method in the field of artificial intelligence and machine learning is originally used to analyze the experimental results of arc fault in the photovoltaic system. The function between the probability of the arc fault and the change of the characteristic variables is obtained. After validating 80 groups of experimental data under different conditions, the accuracy rate of the arc fault detection by this algorithm is proved to reach 100%.

关键词: photovoltaic| arc fault| fast Fourier transform| logistic regression

CLC Number:

TM 914

JIA Fan (贾帆), LUO Liwen *(罗利文), GAO Shiyue (高诗悦), YE Jian (叶健). Logistic Regression Based Arc Fault Detection in Photovoltaic Systems Under Different Conditions[J]. Journal of Shanghai Jiao Tong University (Science), 2019, 24(4): 459-470.

References

[1] National Fire Protection Association. National electricalcode: NFPA 70 [S]. Quincy, MA, USA: NFPA,2011. [2] Underwriters Laboratories Inc. Outline of investigationfor photovoltaic (PV) DC arc-fault circuit protection:UL 1699B [S]. Northbrook, IL, USA: UL, 2013. [3] NOVAK B. Implementing arc detection in solarapplications: achieving compliance with the newUL 1699B standard [EB/OL]. (2012-07-10) [2018-10-10]. http://www.ti.com/lit/wp/spry209/spry209.pdf. [4] XIONG Q, JI S C, ZHU L Y, et al. A novel DC arcfault detection method based on electromagnetic radiationsignal [J]. IEEE Transactions on Plasma Science,2017, 45(3): 472-478. [5] XIA K, HE Z H, YUAN Y, et al. An arc fault detectionsystem for the household photovoltaic inverteraccording to the DC bus currents [C]//18th InternationalConference on Electrical Machines and Systems.Pattaya, Thailand: IEEE, 2015: 1687-1690. [6] CHAE S, PARK J, OH S. Series DC arc fault detectionalgorithm for DC microgrids using relative magnitudecomparison [J]. IEEE Journal of Emerging andSelected Topics in Power Electronics, 2016, 4(4): 1270-1278. [7] CHEN S L, LI X W, XIONG J Y. Series arc faultidentification for photovoltaic system based on timedomainand time-frequency-domain analysis [J]. IEEEJournal of Photovoltaics, 2017, 7(4): 1105-1114. [8] WANG Z, BALOG R S. Arc fault and flash detectionin photovoltaic systems using wavelet transformand support vector machines [C]//43rd IEEE PhotovoltaicSpecialists Conference. Portland, OR, USA:IEEE, 2016: 3275-3280. [9] ZHU H Z, WANG Z, BALOG R S. Real time arc faultdetection in PV systems using wavelet decomposition[C]//43rd IEEE Photovoltaic Specialists Conference.Portland, OR, USA: IEEE, 2016: 1761-1766. [10] LU S B, PHUNG B T, ZHANG D M. A comprehensivereview on DC arc faults and their diagnosis methodsin photovoltaic systems [J]. Renewable and SustainableEnergy Reviews, 2018, 89: 88-98. [11] GUO Y M, WANG L, WU Z Q, et al. Wavelet packetanalysis applied in detection of low-voltage DC arcfault [C]// 4th IEEE Conference on Industrial Electronicsand Applications. Xi’an, China: IEEE, 2009:4013-4016. [12] MENG Z, WANG L, SUN Q G. The characteristics ofDC arc faults current [C]//15th European Conferenceon Power Electronics and Applications. Lille, France:IEEE, 2013: 1-9. [13] YAO X, HERRERA L, JI S C, et al. Characteristicstudy and time-domain discrete-wavelet-transformbased hybrid detection of series DC arc faults [J]. IEEETransactions on Power Electronics, 2014, 29(6): 3103-3115. [14] MOMOH J A, BUTTON R. Design and analysis ofaerospace DC arcing faults using fast Fourier transformationand artificial neural network [C]//IEEEPower Engineering Society General Meeting. Toronto,Canada: IEEE, 2003: 788-793. [15] ROGERS S, GIROLAMI M. A first course in machinelearning [M]. Boca Raton, FL, USA: Chapman andHall/CRC, 2011.