基于特征优化与机器学习的光伏系统轻量化直流串联电弧故障检测方法

doi:10.16628/j.cnki.2095-8188.2026.07.001

摘要/Abstract

摘要：

光伏系统中由连接器松动、电缆绝缘损坏导致的直流串联电弧故障(SAF)因其致火风险高而备受关注。针对光伏系统中直流SAF检测精度低、部署困难等问题,提出一种基于特征优化与机器学习(ML)的光伏系统轻量化直流SAF检测方法。首先利用6级Daubechies-4小波变换结合时域分析,提取出包含44个时频特征的混合特征集。然后比较4种优化策略,采用极端梯度提升(XGBoost)特征重要性(XGB_FI)将特征精简至12个关键指标,经Optuna优化后的XGBoost模型在10折交叉验证中达到99.94%的准确率。最后将轻量化模型部署于树莓派4B进行测试。实验结果表明,模型在4 000个未知样本上准确率仍达99.90%,计算时间可达3.07 ms,满足UL 1699B—2018 Standard for safety for photovoltaic (PV) DC arc-fault circuit protection标准的要求,且所提方法能有效识别SAF。

关键词: 光伏系统, 直流串联电弧故障检测, 特征优化, 轻量化机器学习模型, 嵌入式实时系统

Abstract:

DC series arc faults (SAFs) in photovoltaic (PV) systems are a significant concern due to potential causes such as loose connectors and damaged cable insulation.These faults pose serious fire risks and can lead to increased operational costs.A lightweight machine learning (ML) framework for real-time SAF detection is proposed,with feature optimization adopted to balance detection accuracy and computational efficiency.A hybrid set of 44 time-frequency features is extracted via a 6-level Daubechies-4 wavelet transform combined with time-domain analysis.Four different optimization strategies are evaluated,with XGBoost feature importance (XGB_FI) showing superior performance by reducing the features to 12 key indicators.The Optuna-optimized XGBoost model achieved 99.94% accuracy during 10-fold cross-validation.When deployed on a Raspberry Pi 4B,it achieved 99.90% accuracy on 4 000 unseen samples,with an inference time of just 3.07 ms per sample,which meets the latency requirments specified in UL 1699B—2018 standard for safety for photovoltaic (PV) DC arc-fault circuit protection latency requirements.Effective identification of SAF can be realized by the proposed method.

Key words: photovoltaic system, DC series arc fault detection, feature optimization, lightweight machine learning, embedded real-time system

中图分类号:

TM615

王尧, Md Shoriful Islam, 刘家辉, 盛德杰, 张子哲. 基于特征优化与机器学习的光伏系统轻量化直流串联电弧故障检测方法[J]. 电器与能效管理技术, 2026, 0(7): 1-12.

Wang Yao, Md Shoriful Islam, Liu Jiahui, Sheng Dejie, Zhang Zizhe. Lightweight DC Series Arc Fault Detection in PV Systems Using Machine Learning with Feature Optimization[J]. LOW VOLTAGE APPARATUS, 2026, 0(7): 1-12.

图/表 17

图1

表1

图2

表2

图3

表3

表4

图4

图5

表5

表6

图6

表7

表8

图7

图8

表9

参考文献 28

[1]	IEA. IEA sees great potential for solar,providing up to a quarter of world electricity by 2050[EB/OL].(2010-05-10)[2026-06-08]. https://www.iea.org/news/iea-sees-great-potential-for-solar-providing-up-to-a-quarter-of-world-electricity-by-2050.
[2]	Rystad Energy. China’s solar capacity surges;expected to top 1 TW by 2026[EB/OL].(2023-09-26)[2024-06-26]. https://www.rystadenergy.com/news/china-s-solar-capacity-surges-expected-to-top-1-tw-by-2026.
[3]	王小宇, 刘波, 孙凯, 等. 光伏阵列故障诊断技术综述[J]. 电工技术学报, 2024, 39(20):1-18.
[4]	高少彬, 竺红卫. 直流故障电弧检测技术综述[J]. 电器与能效管理技术, 2018(10):20-24.
[5]	熊庆, 陈维江, 汲胜昌, 等. 低压直流系统故障电弧特性、检测和定位方法研究进展综述[J]. 中国电机工程学报, 2020, 40(18):6015-6027.
[6]	杨庆, 齐玥, 韦思宇, 等. 架空配电线路故障电弧的电磁辐射特性及故障定位应用[J]. 中国电机工程学报, 2024, 44(1):353-362.
[7]	王小君, 窦嘉铭, 刘曌, 等. 可解释人工智能在电力系统中的应用综述与展望[J]. 电力系统自动化, 2024, 48(4):169-191.
[8]	王尧, 马桐桐, 赵宇初, 等. 基于电磁辐射时延估计的串联光伏直流电弧故障定位方法[J]. 电工技术学报, 2023, 38(8):2233-2243.
[9]	Wang L, Lodhi E, Yang P, et al. Adaptive local mean decomposition and multiscale-fuzzy entropy-based algorithms for the detection of DC series arc faults in PV systems[J]. Energies, 2020, 15(10):3608. doi: 10.3390/en15103608
[10]	Li W, Li T, Su J. DC series arc fault detection method for photovoltaic array based on random forest[C]//2024 IEEE 2nd International Conference on Control,Electronics and Computer Technology (ICCECT).Jilin,China,2024:138-143.
[11]	Dang H L, Kwak S G, Choi S. DC series arc fault diagnosis scheme based on hybrid time and frequency features using artificial learning models[J]. Machines, 2024, 12(2):102.1-102.20. doi: 10.3390/machines12010001
[12]	Rufato R C, Ditchi T, Van De Steen C, et al. DC serial arc fault recognition in aircraft using machine learning techniques[J]. International Journal of Electrical Power & Energy Systems, 2025, 164:110408. doi: 10.1016/j.ijepes.2024.110408
[13]	Chen S, Meng Y, Xie Z, et al. Feature selection and detection method of weak arc faults in photovoltaic systems with strong noises based on stochastic resonance[J]. IEEE Transactions on Instrumentation and Measurement, 2022, 71:1-13.
[14]	Yan J, Li Q, Duan S. A simplified current feature extraction and deployment method for DC series arc fault detection[J]. IEEE Transactions on Industrial Electronics, 2024, 71(1):625-634. doi: 10.1109/TIE.2023.3247721
[15]	Song L, Lu C, LI C, et al. Arc detection of photovoltaic DC faults based on mathematical morphology[J]. Machines, 2024, 12(2):134. doi: 10.3390/machines12020134
[16]	Wai R, Cai X. Newly-designed DC-arc fault detection strategy via EMD signal analysis and GRU neural network for solar PV power generation system[C]//5th International Conference on Electrical,Control and Instrumentation Engineering.Kuala Lumpur,Malaysia,2023:1-8.
[17]	王尧, 盛德杰, 李腾, 等. 低压直流串联故障电弧研究综述[J]. 电器与能效管理技术, 2024(8):11-20.
[18]	高汉林, 许峰, 张涵. 光伏系统直流故障电弧特性与检测装置研究[J]. 电器与能效管理技术, 2022(11):46-53.
[19]	刘树鑫, 刘学识, 李静, 等. 基于SSA-ELM的直流串联故障电弧检测方法研究[J]. 电器与能效管理技术, 2022(10):65-73.
[20]	郑力, 鲍光海. 基于改进MobileNet的串联电弧故障检测方法[J]. 电器与能效管理技术, 2024(2):13-20.
[21]	朱海. 基于鲸鱼优化算法改进随机森林的电弧故障检测方法[J]. 电器与能效管理技术, 2024(2):21-27.
[22]	王兆锐, 何键涛, 李治彤, 等. 基于轻量型CNN和特征阈值的光伏系统串联电弧故障检测装置[J]. 电器与能效管理技术, 2024(8):77-85.
[23]	UL 1699B—2018.Standard for safety for photovoltaic (PV) DC arc-fault circuit protection[S].
[24]	Torrence C, Compo G P. A practical guide to wavelet analysis[J]. Bulletin of the American Meteorological Society, 1998, 79(1):61-78. doi: 10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2
[25]	Liaw A, Wiener M. Classification and regression by Random Forest[J]. R News, 2002, 2(3):18-22.
[26]	Chen T, Guestrin C. XGBoost:A scalable tree boosting system[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.San Francisco,CA,USA,2016:785-794.
[27]	Jolliffe I T. Principal component analysis[M]//Everitt B S,Howell D C.Encyclopedia of statistics in behavioral science.New York:John Wiley and Sons Ltd,2005:1580-1584.
[28]	Tharwat A, Gaber T, Ibrahim A, et al. Linear discriminant analysis:A detailed tutorial[J]. AI communications, 2017, 30(2):169-190. doi: 10.3233/AIC-170729

参数	数值	参数	数值
C₁/μF	20	L₂,L₃/μH	60
C₂,C₃/nF	22	L₄,L₅/μH	50
C₄	150 nF,10 μF	L₆,L₇/μH	56
C₅,C₆/nF	1	R₁,R₂/Ω	0
L₁/mH	12	R₃,R₄/Ω	1.0
		R₅,R₆/Ω	0.8

测试编号	最小电弧电流/A	最大功率点电流/A	分离速率/(mm·s^-1)	最大功率点电压/V	开路电压/V	总电阻/Ω	间隙/mm
1	2.5	3	2.5	318	390	56.0	0.8
2	4.4	5	5.0	318	390	21.0	0.8
3	7.0	8	5.0	318	390	21.0	0.8
4	14.0	16	5.0	480	607	11.0	2.5
5	17.0	20	5.0	318	390	8.4	1.3

域名	功能类型	具体特征	数量
小波	统计特征	能量、熵、标准差、峰度、偏度(适用于D₁~D₆、A₆)	30
小波	能量比值	WER_高,WER_中,WER_低	3
时域	形状因子	零交叉率,波峰因数,功率因数	2
时域	迁移统计	局部均值,局部标准差	9
总功能			44

模型	准确率/%	推理时间/ms	训练时间/s
XGBoost	99.93	1.139	0.62
RF	99.82	2.343	4.54
SVC	99.87	0.196	4.86
AdaBoost	99.85	28.137	15.36
GB	99.90	0.431	31.56
Voting (Hard)	99.88	98.741	57.05
Voting (Soft)	99.88	97.655	52.70
Stacking	99.88	98.219	267.74

数据集	总样本数	正常样本数	电弧故障样本数
训练集(80%)	26 868	12 674	14 194
测试集(20%)	6 717	3 169	3 548
总计	33 585	15 843	17 742