基于边缘计算和颜色编码的非侵入式电动自行车充电监测

doi:10.16628/j.cnki.2095-8188.2026.06.002

摘要/Abstract

摘要：

为减少电动自行车违规入户充电引发的火灾事故,非侵入式负荷识别已成为电动自行车监测研究趋势,然而低成本的产品实际应用问题亟待解决。为此,通过电压和傅里叶变换滤波后的电流信号形成U-I轨迹并进行颜色编码处理。运用ShuffleNetV2神经网络实现电动自行车违规室内充电行为的投切检测与识别。同时,为解决实际应用问题,开发了电动自行车充电识别装置。试验表明,所设计的监测装置能够在边缘端准确识别不同品牌电动自行车负荷违规入户充电行为,识别准确率可达98.75%,在不依赖数据传输、避免延时、保护隐私和节省成本的情况下,有效预防火灾事故发生。

关键词: 非侵入式负荷识别, U-I轨迹, 颜色编码, ShuffleNetV2神经网络, K210芯片, 边缘端

Abstract:

To reduce fire accidents induced by illegal indoor charging of electric bicycles, non-intrusive load monitoring has become a mainstream research direction for electric bicycle monitoring. Nevertheless, practical application problems of low-cost monitoring terminals remain to be solved urgently. U-I trajectories are constructed with voltage signals and Fourier-filtered current signals, and color coding is implemented for trajectory feature processing. The ShuffleNetV2 neural network is adopted to realize switching detection and identification of illegal indoor charging behavior of electric bicycles. Meanwhile, an on-site charging identification terminal for electric bicycles is developed to resolve engineering application constraints. Experimental results demonstrate that the proposed monitoring equipment can accurately identify illegal indoor charging loads of various electric bicycle brands at the edge computing terminal, with an identification accuracy of 98.75%. It can effectively prevent fire hazards without data transmission dependence, transmission delay, privacy leakage and extra hardware cost.

Key words: non-intrusive load monitoring, U-I trajectory, color coding, ShuffleNetV2 neural network, K210 chip, edge computing

中图分类号:

TM714

汪志强, 鲍光海, 孟立新. 基于边缘计算和颜色编码的非侵入式电动自行车充电监测[J]. 电器与能效管理技术, 2026, 0(6): 7-17.

WANG Zhiqiang, BAO Guanghai, MENG Lixin. Non-Intrusive Charging Monitoring of Electric Bicycles Based on Edge Computing and Color Coding[J]. LOW VOLTAGE APPARATUS, 2026, 0(6): 7-17.

图/表 18

图1

图2

图3

图4

图5

图6

图7

表1

表2

表3

图8

表4

图9

图10

表5

图11

表6

表7

参考文献 20

[1]	LI M, LI Z. A hybrid event detection approach for non-intrusive load monitoring[J]. IEEE Transactions on Smart Grid, 2020, 11(1):528-540. doi: 10.1109/TSG.2019.2924862
[2]	RUZZELLI A G, NICOLAS C, SCHOOFS A, et al. Real-time recognition and profiling of appliances through a single electricity sensor[C]// 2010 7th Annual IEEE Communications Society Conference on Sensor,Mesh and Ad Hoc Communications and Networks (SECON). Boston,MA,USA,2010:1-9.
[3]	牛卢璐, 贾宏杰. 一种适用于非侵入式负荷监测的暂态事件检测算法[J]. 电力系统自动化, 2011, 35(9):30-35.
[4]	李想, 刘宇航, 张琪, 等. 小波检测和特征图谱决策的非侵入电动自行车充电实时监测系统[J]. 电力系统自动化, 2023, 47(19):177-186.
[5]	鲍光海, 黄逸欣. 基于ResNeXt网络和迁移学习的非侵入式负荷监测[J]. 电力系统自动化, 2023, 47(13):110-120.
[6]	周明, 宋旭帆, 涂京, 等. 基于非侵入式负荷监测的居民用电行为分析[J]. 电网技术, 2018, 42(10):3268-3276.
[7]	崔昊杨, 蔡杰, 陈磊, 等. 基于颜色编码的非侵入式负荷细粒度识别方法[J]. 电网技术, 2022, 46(4):1557-1567.
[8]	魏广芬, 赵航, 胡春华, 等. 基于香农熵加权投票算法的集成式非侵入式负荷识别方法[J]. 中国电机工程学报, 2022, 42(24):8876-8888.
[9]	孟立新, 汪志强, 鲍光海. 基于时空特征编码与动态分类器扩展的非侵入式未知负荷增量识别[J/OL]. 中国电机工程学报,1-16 (2025-12-01)[2025-12-17].https://doi.org/10.13334/j.0258-8013.pcsee.251301.
[10]	魏星琦, 张宸宇, 缪惠宇, 等. 基于多特征融合的非侵入式负荷辨识方法研究[J]. 电器与能效管理技术, 2023(7):55-60.
[11]	段佳其, 鲍光海, 方艳东. 基于V-I轨迹的非侵入式电动自行车充电行为在线辨识[J]. 电器与能效管理技术, 2024(12):69-76.
[12]	施雨松, 徐青山, 郑建. 基于特征选择与增量学习的非侵入式电动自行车充电辨识方法[J]. 电力系统自动化, 2021, 45(7):87-94.
[13]	栾文鹏, 马纯伟, 赵博超, 等. 非侵入式电动自行车充电负荷在线识别[J]. 电力系统及其自动化学报, 2022, 34(10):1-7.
[14]	LIU Y, XU Q, YANG Y, et al. Detection of electric bicycle indoor charging for electrical safety:A NILM approach[J]. IEEE Transactions on Smart Grid, 2023, 14(5):3862-3875. doi: 10.1109/TSG.2023.3245636
[15]	赵杰, 张延平, 丁媛媛. 一种电动自行车充电负荷波形特征提取及识别方法[J]. 电器与能效管理技术, 2025(8):25-33.
[16]	LE T T H, KANG H, KIM H L. Household appliance classification using lower odd-numbered harmonics and the bagging decision tree[J]. IEEE Access, 2020, 8(1):55937-55952. doi: 10.1109/Access.6287639
[17]	HASSAN T, JAVED F, ARSHAD N. An empirical investigation of V-I trajectory based load signatures for non-intrusive load monitoring[C]// 2014 IEEE PES General Meeting,Conference & Exposition.National Harbor,MD, USA,2014:870-878.
[18]	HE D, DU L, YANG Y, et al. Front-end electronic circuit topology analysis for model-driven classification and monitoring of appliance loads in smart buildings[J]. IEEE Transactions on Smart Grid, 2012, 3(4):2286-2293. doi: 10.1109/TSG.2012.2219327
[19]	MA N, ZHANG X, ZHENG H, et al. ShuffleNet V2:Practical guidelines for efficient CNN architecture design[J]. Lecture Notes in Computer Science, 2018, 11218(1):122-138.
[20]	ZHANG X, ZHOU X, LIN M, et al. ShuffleNet:An extremely efficient convolutional neural network for mobile devices[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City,UT, USA,2018:6848-6856.

层名称	输出大小	卷积核大小	重复次数
Image	120×120×3	—	—
Conv1	60×60×12	3×3	1
MaxPool	30×30×12	3×3	1
Stage2	15×15×24	—	2
Stage3	8×8×48	—	4
Stage4	4×4×96	—	2
Conv5	4×4×512	1×1	1
GlobalPool	1×1×512	4×4	1
FC	1×1×2	—	1

简称	品牌	规格
EB1	雅迪	60 V 20 Ah
EB2	小刀	60 V 20 Ah
EB3	欧派特	60 V 20 Ah
EB4	雅迪	48 V 20 Ah
EB5	爱玛	48 V 20 Ah
EB6	雅迪	48 V 12 Ah
EB7	立马	48 V 12 Ah
EB8	爱玛	48 V 12 Ah

运行负荷	投切负荷	TP	FN	TN
空载	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	10	0	0
	EB8	10	0	0
	非EB	0	0	50
微波炉	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	9	1	0
	EB8	10	0	0
	非EB	0	0	50
吹风机、笔记本电脑、台灯、风扇	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	10	0	0
	EB8	10	0	0
	非EB	0	0	50
其他	EB	247	3	0
其他	非EB	0	0	250
总计		396	4	400

模型名称	准确率/%	模型大小/KB	FLOPs	参数量
EfficientNet-lite	99.75	5 601	5.781 0×10⁷	1.400 0×10⁶
SqueezeNet	98.00	2 910	1.984 4×10⁸	7.364 5×10⁵
MobileNetV3	99.50	998	3.4300×10⁶	2.306 8×10⁵
MobileNetV2	99.25	799	1 236 0×10⁷	1.600 3×10⁵
ShuffleNetV2	99.50	377	3.550 0×10⁶	7.331 0×10⁴

运行负荷	投切负荷	TP	FN	TN
空载	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	10	0	0
	EB8	10	0	0
	非EB	0	0	50
微波炉	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	8	2	0
	EB8	9	1	0
	非EB	0	0	50
吹风机、笔记本电脑、台灯、风扇	EB2	10	0	0
	EB3	10	0	0
	EB5	10	0	0
	EB7	10	0	0
	EB9	10	0	0
	非EB	0	0	50
其他	EB	244	7	0
其他	非EB	0	0	250
总计		390	10	400