False data injection attacks detection model based on LightGBM for household load data

doi:10.3969/j.issn.2097-0706.2024.11.001

Abstract

Abstract:

With the continuous development of information technology， smart meters have been widely deployed in many households， allowing power companies to better identify the socio-demographic characteristics of power consumers and provide diversified services. However， one of the threats faced by smart grids is energy theft， particularly through False Data Injection Attacks （FDIA）， which tamper with meter data for covert theft， posing a serious threat to the safe and stable operation of power systems. To address this issue， an FDIA detection model based on LightGBM was proposed. Normal user electricity consumption data was selected， and different types of FDIA were implemented on some users. Features were extracted using a sliding window method， and the LightGBM model was employed for multi-class detection. Experimental results showed that this model excelled in detection accuracy and real-time performance， accurately identifying different types of FDIA with quick and efficient detection， meeting the real-time requirements of practical applications. This model could help ensure the safe operation of power systems.

Key words: smart meter, false data injection attack, detection model, feature extraction, LightGBM, data analysis

CLC Number:

TK01：TM721

WANG Jin, ZHANG Xiaoyu. False data injection attacks detection model based on LightGBM for household load data[J]. Integrated Intelligent Energy, 2024, 46(11): 1-9.

Figures/Tables 11

Table 1

Different types of FDIA

攻击类型	修改方式
FDIA1	$x ˜ t ← α x t$ , 其中， $0.2 < α < 0.8$ 随机生成， $t$ 时段内所有 $α$ 相同
FDIA2	$x ˜ t ← m a x x t - γ, 0$ , 其中， $γ$ 为常数，从 $0, m a x x$ 中随机生成
FDIA3	$x ˜ t ← x t x t ≤ γ γ 其他,$ 其中， $γ$ 为随机的截止点， $γ < m a x x$ 为标准偏差
FDIA4	$x ˜ t ← α t x t$ , 其中， $0.2 < α < 0.8$ 随机生成， $t$ 时段内每个 $α$ 都不同
FDIA5	$x ˜ t ← α t x -$ , 其中， $0.2 < α < 0.8$ 随机生成， $t$ 时段内每个 $α$ 都不同
FDIA6	$x ˜ t ← f t x t$ , $f t = 0 t 1 < t < t 2 1 其他$ , 其中， $t$ 为随机生成的时间段，该时间段大于4 h

Table 1

Fig.1

Fig.2

Table 2

120 features extracted from weekly data

特征集	特征值
统计特征（60个）	每周（工作日、周末）平均负荷每周（工作日、周末）最大负荷每周（工作日、周末）最小负荷每周（工作日、周末）总负荷每周（工作日、周末）基准（早、中、晚）平均电荷每周（工作日、周末）基准（早、中、晚）最大电荷每周（工作日、周末）基准（早、中、晚）最小电荷每周（工作日、周末）基准（早、中、晚）总电荷
比率特征（38个）	负载因子（7 d）每周（工作日、周末）基准（早、中、晚）负载因子每周（工作日、周末）（最小负荷/平均负荷）每周（工作日、周末）（平均负荷/最大负荷）每周（早、中、晚）平均负荷/每周（工作日、周末）基准负荷每周（工作日、周末）平均负荷/每周平均负荷每周（工作日、周末）总负荷/每周总负荷
分布特征（22个）	$P x ≥ 0.25 k W, P x ≥ 0.50 k W, P x ≥ 1.00 k W, P x ≥ 2.00 k W, P x ≥ x -$ 第1个最大（最小）负荷出现的位置最后一个最大（最小）负荷出现的位置每周的方差每周第25（50，75）百分位数每周的偏度每周的峰度每周的熵自相关函数（间隔1—6 d）

Table 2

Fig.3

Table 3

Table 4

Table 5

Fig.4

Fig.5

Table 6

References 27

[1]	张利生. 电力网电能损耗管理及降损技术[M]. 2版. 北京: 中国电力出版社, 2008.
[2]	GAO Y Q, FOGGO B, YU N P. A physically inspired data-driven model for electricity theft detection with smart meter data[J]. IEEE Transactions on Industrial Informatics, 2019, 15(9): 5076-5088.
[3]	RAMOS C C O, RODRIGUES D, DE SOUZA A N, et al. On the study of commercial losses in Brazil: A binary black hole algorithm for theft characterization[J]. IEEE Transactions on Smart Grid, 2018, 9(2): 676-683.
[4]	NAGI J, YAP K S, NAGI F, et al. NTL detection of electricity theft and abnormalities for large power consumers in TNB Malaysia[C]// Proceedings of 2010 IEEE Student Conference on Research and Development(SCOReD). IEEE, 2010: 202-206.
[5]	LIU Y, NING P, REITER M K. False data injection attacks against state estimation in electric power grids[J]. ACM Transactions on Information and System Security, 2011, 14(1): 1-33.
[6]	陈震宇, 关志涛. 基于ARIMA模型的智能电网虚假数据注入攻击检测[J]. 电力信息与通信技术, 2021, 19(11): 24-29.
	CHEN Zhenyu, GUAN Zhitao. ARIMA model-based false data injection attack detection in smart grid[J]. Electric Power Information and Communication Technology, 2021, 19(11): 24-29.
[7]	陈冰, 唐永旺. 基于Bi-GRU和自注意力的智能电网虚假数据注入攻击检测[J]. 计算机应用与软件, 2021, 38(7): 339-344, 349.
	CHEN Bing, TANG Yongwang. False data injection attacks detection in smart grid based on Bi-GRU and self-attention[J]. Computer Applications and Software, 2021, 38(7): 339-344, 349.
[8]	黄冬梅, 王一帆, 胡安铎, 等. 融合无监督和有监督学习的虚假数据注入攻击检测[J]. 电力工程技术, 2024, 43(2): 134-141.
	HUANG Dongmei, WANG Yifan, HU Anduo, et al. Detection method of false data injection attack based on unsupervised and supervised learning[J]. Electric Power Engineering Technology, 2024, 43(2): 134-141.
[9]	胡聪, 洪德华, 张翠翠, 等. 一种基于特征映射与深度学习的虚假数据注入检测方法[J]. 现代电力, 2023, 40(1): 125-132.
	HU Cong, HONG Dehua, ZHANG Cuicui, et al. A method to detect false data injection based on feature mapping and deep learning[J]. Modern Electric Power, 2023, 40(1): 125-132.
[10]	陈冰, 唐永旺. 基于Transformer编码器的智能电网虚假数据注入攻击检测[J]. 计算机应用与软件, 2022, 39(7): 336-342.
	CHEN Bing, TANG Yongwang. False data injection attacks detecting based on transformer encoder in smart grid[J]. Computer Applications and Software, 2022, 39(7): 336-342.
[11]	赵峰, 李妞妞. 基于VAE-GWO-LightGBM的信用卡欺诈检测方法[J]. 东北师大学报(自然科学版), 2023, 55(4): 77-84.
	ZHAO Feng, LI Niuniu. Credit card fraud detection method based on VAE-GWO-LightGBM[J]. Journal of Northeast Normal University(Natural Science Edition), 2023, 55(4): 77-84.
[12]	王鑫, 廖彬, 李敏, 等. 融合LightGBM与SHAP的糖尿病预测及其特征分析方法[J]. 小型微型计算机系统, 2022, 43(9): 1877-1885.
	WANG Xin, LIAO Bin, LI Min, et al. Combination of LightGBM and SHAP for diabetes prediction and feature analysis[J]. Journal of Chinese Computer Systems, 2022, 43(9): 1877-1885.
[13]	王懿泽, 孙吉利, 闫成杰, 等. 基于超像素与LightGBM的极化SAR图像地物分类[J]. 中国科学院大学学报, 2023, 40(5): 658-669.
	WANG Yize, SUN Jili, YAN Chengjie, et al. Polarimetric SAR image terrain classification based on superpixel and LightGBM[J]. Journal of University of Chinese Academy of Sciences, 2023, 40(5): 658-669.
[14]	肖宏磊, 留毅, 夏红军, 等. 基于PCA与SSA-LightGBM的油浸式变压器故障诊断方法[J]. 综合智慧能源, 2023, 45(3): 9-16. doi: 10.3969/j.issn.2097-0706.2023.03.002
	XIAO Honglei, LIU Yi, XIA Hongjun, et al. Oil-immersed transformer fault diagnosis method based on PCA and SSA-LightGBM[J]. Integrated Intelligent Energy, 2023, 45(3): 9-16. doi: 10.3969/j.issn.2097-0706.2023.03.002
[15]	秦宇宸. 大数据技术在反窃电中的应用[J]. 光源与照明, 2022(7):195-197.
	QIN Yuchen. The application of big data technology in the field of anti-electricity stealing—Taking a line with high line loss rate as an example[J]. Lamps & Lighting, 2022(7):195-197.
[16]	马晓杰, 程晓荣. 智能电网中的网络攻击检测机制的研究[J]. 网络安全技术与应用, 2020(6):28-30.
	MA Xiaojie, CHENG Xiaorong. Research on network attack detection mechanism in smart grid[J]. Network Security Technology and Application, 2020(6):28-30.
[17]	李惠军, 陆建强, 周霞, 等. 面向智慧园区系统的网络攻击关联分析与防护策略研究[J]. 综合智慧能源, 2022, 44(7): 1-9. doi: 10.3969/j.issn.2097-0706.2022.07.001
	LI Huijun, LU Jianqiang, ZHOU Xia, et al. Network attack association analysis and attack protection strategy for smart park systems[J]. Integrated Intelligent Energy, 2022, 44(7): 1-9. doi: 10.3969/j.issn.2097-0706.2022.07.001
[18]	鲁杰, 杨超, 杜刃刃, 等. 电力CPS中的虚假数据注入攻击[J]. 智能计算机与应用, 2022, 12(6): 121-126.
	LU Jie, YANG Chao, DU Renren, et al. False data injection attack in power CPS[J]. Intelligent Computer and Applications, 2022, 12(6): 121-126.
[19]	ZANETTI M, JAMHOUR E, PELLENZ M, et al. A tunable fraud detection system for advanced metering infrastructure using short-lived patterns[J]. IEEE Transactions on Smart Grid, 2019, 10(1): 830-840.
[20]	ELAHE M F, JIN M, ZENG P. Knowledge-based systematic feature extraction for identifying households with plug-in electric vehicles[J]. IEEE Transactions on Smart Grid, 2022, 13(3): 2259-2268.
[21]	WANG Y, BENNANI I L, LIU X F, et al. Electricity consumer characteristics identification: A federated learning approach[J]. IEEE Transactions on Smart Grid, 2021, 12(4): 3637-3647.
[22]	石洪波, 陈雨文, 陈鑫. SMOTE过采样及其改进算法研究综述[J]. 智能系统学报, 2019, 14(6): 1073-1083.
	SHI Hongbo, CHEN Yuwen, CHEN Xin. Summary of research on SMOTE oversampling and its improved algorithms[J]. CAAI Transactions on Intelligent Systems, 2019, 14(6): 1073-1083.
[23]	汪李忠, 池建飞, 丁叶强, 等. 基于NNTR-SMOTE与GA-XGBoost的变压器故障诊断方法研究[J]. 综合智慧能源, 2024, 46(1): 84-93. doi: 10.3969/j.issn.2097-0706.2024.01.010
	WANG Lizhong, CHI Jianfei, DING Yeqiang, et al. Transformer fault diagnosis method based on NNTR-SMOTE and GA-XGBoost[J]. Integrated Intelligent Energy, 2024, 46(1): 84-93. doi: 10.3969/j.issn.2097-0706.2024.01.010
[24]	KE G L, MENG Q, FINLEY T, et al. LightGBM: A highly efficient gradient boosting decision tree[C]// International Conference on Neural Information Processing Systems. NeurIPS, 2017 : 3146-3154.
[25]	丁建立, 孙玥. 基于LightGBM的航班延误多分类预测[J]. 南京航空航天大学学报, 2021, 53(6): 847-854.
	DING Jianli, SUN Yue. Multi-classification prediction of flight delay based on LightGBM[J]. Journal of Nanjing University of Aeronautics & Astronautics, 2021, 53(6): 847-854.
[26]	蒋其容, 魏勇, 高先松, 等. LSTM-LightGBM组合模型的短期电力负荷预测[J]. 中国设备工程, 2023(8): 78-81.
	JIANG Qirong, WEI Yong, GAO Xiansong, et al. Short-term power load forecasting based on LSTM-LightGBM combined model[J]. China Plant Engineering, 2023(8): 78-81.
[27]	SCHOFIELD J R, TINDEMANS S, CARMICHAEL R, et al. Low Carbon London project: Data from the dynamic time-of-use electricity pricing trial,2013[EB/OL].(2015-11-01)[2024-05-01].http://doi.org/10.5255/ UKDA-SN-7857-1.

特征集	特征值
统计特征（60个）	每周（工作日、周末）平均负荷每周（工作日、周末）最大负荷每周（工作日、周末）最小负荷每周（工作日、周末）总负荷每周（工作日、周末）基准（早、中、晚）平均电荷每周（工作日、周末）基准（早、中、晚）最大电荷每周（工作日、周末）基准（早、中、晚）最小电荷每周（工作日、周末）基准（早、中、晚）总电荷
比率特征（38个）	负载因子（7 d）每周（工作日、周末）基准（早、中、晚）负载因子每周（工作日、周末）（最小负荷/平均负荷）每周（工作日、周末）（平均负荷/最大负荷）每周（早、中、晚）平均负荷/每周（工作日、周末）基准负荷每周（工作日、周末）平均负荷/每周平均负荷每周（工作日、周末）总负荷/每周总负荷
分布特征（22个）	$P x ≥ 0.25 k W, P x ≥ 0.50 k W, P x ≥ 1.00 k W, P x ≥ 2.00 k W, P x ≥ x -$ 第1个最大（最小）负荷出现的位置最后一个最大（最小）负荷出现的位置每周的方差每周第25（50，75）百分位数每周的偏度每周的峰度每周的熵自相关函数（间隔1—6 d）

类别	真实正常用户	真实异常用户
预测正常用户	TP	FP
预测异常用户	FN	TN

数据集	正常样本	攻击样本	全部样本
训练集	22 879	2 618	25 497
测试集	9 799	1 129	10 928

攻击类型	LSTM			CNN-TCN			SVM			LightGBM
攻击类型	R_pre	R_rec	F₁	R_pre	R_rec	F₁	R_pre	R_rec	F₁	R_pre	R_rec	F₁
0	0.36	0.82	0.50	1.00	0.96	0.98	0.99	1.00	0.99	1.00	1.00	1.00
FDIA1	0.71	0.31	0.43	0.98	1.00	0.99	0.96	0.94	0.95	1.00	0.99	1.00
FDIA2	0.95	0.97	0.96	0.86	0.98	0.91	0.97	0.89	0.92	0.99	1.00	1.00
FDIA3	0.98	0.77	0.87	0.92	0.84	0.88	0.93	0.97	0.95	1.00	1.00	1.00
FDIA4	1.00	0.92	0.96	0.86	0.92	0.89	1.00	0.93	0.96	1.00	0.95	0.98
FDIA5	0.98	1.00	0.99	1.00	0.85	0.92	0.98	0.99	0.99	0.97	0.99	0.98
FDIA6	0.90	0.52	0.66	0.95	1.00	0.98	0.96	0.53	0.68	1.00	0.98	0.99

攻击类型	统计特征			比率特征			分布特征			所有特征
攻击类型	R_pre	R_rec	F₁	R_pre	R_rec	F₁	R_pre	R_rec	F₁	R_pre	R_rec	F₁
0	0.96	1.00	0.98	0.96	1.00	0.98	1.00	1.00	1.00	1.00	1.00	1.00
FDIA1	0.00	0.00	0.00	0.00	0.00	0.00	1.00	0.99	1.00	1.00	0.99	1.00
FDIA2	0.95	0.76	0.85	0.97	0.82	0.89	0.99	0.98	0.99	0.99	1.00	1.00
FDIA3	1.00	0.94	0.97	0.96	0.77	0.85	1.00	1.00	1.00	1.00	1.00	1.00
FDIA4	0.98	0.75	0.85	0.97	0.88	0.92	0.99	0.96	0.98	1.00	0.95	0.98
FDIA5	0.94	0.97	0.96	0.97	0.91	0.94	0.98	0.98	0.98	0.97	0.99	0.98
FDIA6	0.83	0.46	0.59	0.96	0.46	0.62	0.98	1.00	0.99	1.00	0.98	0.99