基于两阶段数据增强和双向深度残差TCN的用户负荷曲线分类方法

doi:10.12204/j.issn.1000-7229.2022.02.011

电力建设 ›› 2022, Vol. 43 ›› Issue (2): 89-97.doi: 10.12204/j.issn.1000-7229.2022.02.011

基于两阶段数据增强和双向深度残差TCN的用户负荷曲线分类方法

张杰¹(), 刘洋¹^,²(), 李文峰³(), 王磊¹(), 许立雄¹()

1.四川大学电气工程学院,成都市 610065
2.智能电网四川省重点实验室(四川大学),成都市 610065
3.国网河南省电力公司, 郑州市 450052

收稿日期:2021-07-26 出版日期:2022-02-01 发布日期:2022-03-24
通讯作者: 刘洋 E-mail:zj@stu.scu.edu.cn;yang.liu@scu.edu.cn;809406879@qq.com;1503994462@qq.com;xulixiong@163.com
作者简介:张杰(1997),男,硕士研究生,研究方向为电力系统数据挖掘与用电行为精细化辨识,E-mail: zj@stu.scu.edu.cn;
李文峰(1985),男,博士,高级工程师,研究方向为电力市场和电能质量等,E-mail: 809406879@qq.com;
王磊(1995),男,硕士研究生,研究方向为电力系统自动化、人工智能等,E-mail: 1503994462@qq.com;
许立雄(1982),男,博士,副教授,研究方向为人工智能在电力系统分析中的运用、微能源网规划与运行等,E-mail: xulixiong@163.com。
基金资助:
国家电网公司科技项目(5217L021000C)

Power Load Curve Identification Method Based on Two-Phase Data Enhancement and Bi-directional Deep Residual TCN

ZHANG Jie¹(), LIU Yang¹^,²(), LI Wenfeng³(), WANG Lei¹(), XU Lixiong¹()

1. College of Electrical Engineering, Sichuan University, Chengdu 610065, China
2. Key Laboratory of Intelligent Electric Power Grid of Sichuan Province (Sichuan University),Chengdu 610065, China
3. State Grid Henan Electric Power Company, Zhengzhou 450052, China

Received:2021-07-26 Online:2022-02-01 Published:2022-03-24
Contact: LIU Yang E-mail:zj@stu.scu.edu.cn;yang.liu@scu.edu.cn;809406879@qq.com;1503994462@qq.com;xulixiong@163.com
Supported by:
State Grid Corporation of China Research Program(5217L021000C)

摘要/Abstract

摘要：

深度挖掘用户负荷规律并感知用电行为对于提升电网服务质量、改善用户用能体验具有重要意义。针对用户负荷中存在的数据缺失、类别不平衡问题以及分类模型性能缺陷,提出一种基于数据增强和双向深度残差时间卷积网络(temporal convolutional network,TCN)的电力用户负荷曲线分类方法。首先,提出考虑负荷数据全局分布特性的两阶段数据增强方法,第一阶段采用基于张量奇异值阈值算法的低秩张量补全方法补全缺失数据,第二阶段使用基于Wasserstein距离的生成对抗网络过采样少数类样本,解决类别不平衡问题。其次,构建融合双向时序特征的深度残差TCN分类模型,实现大规模用电曲线精准辨识。最后,通过选取公开时序分类基准数据集以及实测负荷数据集,验证了所提分类模型在收敛速度和分类精度上具有更好的性能,所提数据增强方法能有效提升模型分类效果。

关键词: 负荷分类, 时间卷积网络(TCN), 生成对抗网络, 低秩张量补全, 类别不平衡, 数据缺失

Abstract:

Deeply mining power consumer load law and perceiving electricity consumption behavior are of great significance for improving the service quality of power grid and the experience of power consumer. To deal with the issues of missing data, category imbalance and performance defects of classification model, this paper proposes a power user load curve classification method based on data enhancement and improved temporal convolution network (TCN). Firstly, a two-phase data enhancement method is proposed considering the global distribution characteristics of load data. In the first phase, the low rank tensor completion (LRTC) method based on tensor singular value threshold algorithm is introduced to complete the missing data. In the second phase, the generation adversarial network based on Wasserstein distance (WGAN) is used to enhance the minority samples to solve the problem of class imbalance. Secondly, a modified deep TCN classification model integrating bi-directional time-series features is constructed to realize accurate identification of large-scale power consumption curves. Finally, through the open-source criteria temporal dataset for classification and the actual load dataset, the proposed classification model shows better performance in convergence speed and classification accuracy, and the proposed data enhancement method can effectively improve the classification effect of models.

Key words: load classification, temporal convolutional network (TCN), generative adversarial network, low rank tensor completion, class imbalance, data missing

中图分类号:

TM714

张杰, 刘洋, 李文峰, 王磊, 许立雄. 基于两阶段数据增强和双向深度残差TCN的用户负荷曲线分类方法[J]. 电力建设, 2022, 43(2): 89-97.

ZHANG Jie, LIU Yang, LI Wenfeng, WANG Lei, XU Lixiong. Power Load Curve Identification Method Based on Two-Phase Data Enhancement and Bi-directional Deep Residual TCN[J]. ELECTRIC POWER CONSTRUCTION, 2022, 43(2): 89-97.

图/表 12

图1 GAN原理

Fig.1 Schematic diagram of GAN

图2 时间卷积网络模型

Fig.2 Temporal convolutional network

图3 深度残差TCN层

Fig.3 Deep residual blocks of TCN

图4 Bi-DR-TCN模型

Fig.4 Bi-DR-TCN model

图5 负荷分类流程

Fig.5 Load classification framework

表1 UCR数据集参数

Table 1 Parameters of UCR data set

表2 UCR数据实验结果

Table 2 Experiment results on UCR data

图6 训练迭代对比

Fig.6 Comparison of training iterations

图7 缺失补全对比

Fig.7 Comparison between missing and completion

图8 生成样本与原始样本对比

Fig.8 Comparison between synthetic and real samples

图9 负荷分类结果

Fig.9 Load classification results

表3 不同数据增强方法效果对比

Table 3 Comparison of different data enhancement methods

参考文献 26

[1]	陈启鑫, 刘敦楠, 林今, 等. 能源互联网的商业模式与市场机制(一)[J]. 电网技术, 2015, 39(11):3050-3056.
	CHEN Qixin, LIU Dunnan, LIN Jin, et al. Business models and market mechanisms of energy internet(1)[J]. Power System Technology, 2015, 39(11):3050-3056.
[2]	朱天怡, 艾芊, 贺兴, 等. 基于数据驱动的用电行为分析方法及应用综述[J]. 电网技术, 2020, 44(9):3497-3507.
	ZHU Tianyi, AI Qian, HE Xing, et al. An overview of data-driven electricity consumption behavior analysis method and application[J]. Power System Technology, 2020, 44(9):3497-3507.
[3]	王睿, 高欣, 李军良, 等. 基于聚类分析的电动汽车充电负荷预测方法[J]. 电力系统保护与控制, 2020, 48(16):37-44.
	WANG Rui, GAO Xin, LI Junliang, et al. Electric vehicle charging demand forecasting method based on clustering analysis[J]. Power System Protection and Control, 2020, 48(16):37-44.
[4]	陈明照, 毛坚, 杜宗林, 等. 基于聚类法的工业用户需求侧管理(DSM)方案分析与研究[J]. 电力系统保护与控制, 2017, 45(7):84-89.
	CHEN Mingzhao, MAO Jian, DU Zonglin, et al. Analysis on demand side management scheme of industrial enterprise based on clustering method[J]. Power System Protection and Control, 2017, 45(7):84-89.
[5]	王桂兰, 周国亮, 赵洪山, 等. 大规模用电数据流的快速聚类和异常检测技术[J]. 电力系统自动化, 2016, 40(24):27-33.
	WANG Guilan, ZHOU Guoliang, ZHAO Hongshan, et al. Fast clustering and anomaly detection technique for large-scale power data stream[J]. Automation of Electric Power Systems, 2016, 40(24):27-33.
[6]	刘洋, 刘洋, 许立雄. 适用于海量负荷数据分类的高性能反向传播神经网络算法[J]. 电力系统自动化, 2018, 42(21):96-103.
	LIU Yang, LIU Yang, XU Lixiong. High-performance back propagation neural network algorithm for classification of mass load data[J]. Automation of Electric Power Systems, 2018, 42(21):96-103.
[7]	林顺富, 顾乡, 汤继开, 等. 基于稀疏自动编码器神经网络的负荷曲线分类方法[J]. 电网技术, 2020, 44(9):3508-3515.
	LIN Shunfu, GU Xiang, TANG Jikai, et al. Power load profile classification method based on neural network of sparse automatic encoder[J]. Power System Technology, 2020, 44(9):3508-3515.
[8]	彭文, 王金睿, 尹山青. 电力市场中基于Attention-LSTM的短期负荷预测模型[J]. 电网技术, 2019, 43(5):1745-1751.
	PENG Wen, WANG Jinrui, YIN Shanqing. Short-term load forecasting model based on attention-LSTM in electricity market[J]. Power System Technology, 2019, 43(5):1745-1751.
[9]	BAI S, KOLTER J Z, KOLTUN V. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling[DB/OL]. (2018-04-19)[2021-02-12]. https://arxiv.org/abs/1803.01271v1.
[10]	LEA C, FLYNN M D, VIDAL R, et al. Temporal convolutional networks for action segmentation and detection[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017: 1003-1012.
[11]	LEA C, VIDAL R, REITER A, et al. Temporal convolutional networks: A unified approach to action segmentation[M]// Lecture Notes in Computer Science. Cham: Springer International Publishing, 2016: 47-54.
[12]	TUTZ G, RAMZAN S. Improved methods for the imputation of missing data by nearest neighbor methods[J]. Computational Statistics & Data Analysis, 2015, 90:84-99. doi: 10.1016/j.csda.2015.04.009 URL
[13]	王守相, 陈海文, 潘志新, 等. 采用改进生成式对抗网络的电力系统量测缺失数据重建方法[J]. 中国电机工程学报, 2019, 39(1):56-64,320.
	WANG Shouxiang, CHEN Haiwen, PAN Zhixin, et al. A reconstruction method for missing data in power system measurement using an improved generative adversarial network[J]. Proceedings of the CSEE, 2019, 39(1):56-64,320.
[14]	乔文俞, 李野, 刘浩宇, 等. 基于曲线相似与低秩矩阵的缺失电量数据补全方法[J]. 电力建设, 2020, 41(1):32-38.
	QIAO Wenyu, LI Ye, LIU Haoyu, et al. Missing load data completion based on curve similarity and low-rank matrix[J]. Electric Power Construction, 2020, 41(1):32-38.
[15]	刘洋, 刘洋, 许立雄, 等. 计及数据类别不平衡的海量用户负荷典型特征高性能提取方法[J]. 中国电机工程学报, 2019, 39(14):4093-4104.
	LIU Yang, LIU Yang, XU Lixiong, et al. A high performance extraction method for massive user load typical characteristics considering data class imbalance[J]. Proceedings of the CSEE, 2019, 39(14):4093-4104.
[16]	李想, 王鹏, 刘洋, 等. 考虑类别不平衡的海量负荷用电模式辨识方法[J]. 中国电机工程学报, 2020, 40(1):128-137.
	LI Xiang, WANG Peng, LIU Yang, et al. Massive load pattern identification method considering class imbalance[J]. Proceedings of the CSEE, 2020, 40(1):128-137.
[17]	赵洪山, 寿佩瑶, 马利波. 低压台区缺失数据的张量补全方法[J]. 中国电机工程学报, 2020, 40(22):7328-7337.
	ZHAO Hongshan, SHOU Peiyao, MA Libo. A tensor completion method of missing data in transformer district[J]. Proceedings of the CSEE, 2020, 40(22):7328-7337.
[18]	GANDY S, RECHT B, YAMADA I. Tensor completion and low-n-rank tensor recovery via convex optimization[J]. Inverse Problems, 2011, 27(2):025010. doi: 10.1088/0266-5611/27/2/025010 URL
[19]	GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial nets [C]//International Conference on Neural Information Processing Systems. Montreal: MIT Press, 2014:2672-2680.
[20]	ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein GAN[DB/OL].(2017-01-26)[2021-2-12]. https://arxiv.org/abs/1701.07875v3.
[21]	CLEVERT D A, UNTERTHINER T, HOCHREITER S. Fast and accurate deep network learning by exponential linear units(elus)[DB/OL].(2015-12-05)[2021-02012]. https://arxiv.org/pdf/1511.07289v2.pdf.
[22]	SRIVASTAVA N, HINTON G, KRIZHEVSKY A, et al. Dropout: A simple way to prevent neural networks from overfitting[J]. Journal of Machine Learning Research, 2014, 15(6):1929-1958.
[23]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016: 770-778.
[24]	ZHANG S, ZHENG D, HU X, et al. Bidirectional long short- term memory networks for relation classification[C]// Proceedings of the 29th Pacific Asia Conference on Language, Information and Computation. 2015: 73-78.
[25]	HILLS J, LINES J, BARANAUSKAS E, et al. Classification of time series by shapelet transformation[J]. Data Mining and Knowledge Discovery, 2014, 28(4):851-881. doi: 10.1007/s10618-013-0322-1 URL
[26]	FRANK A, ASUNCION A. UCI machine learning repository[EB/OL]. Irvine, USA: University of California, School of Information and Computer Science.(2015-03-19)[2021-03-12]. http://archive.ics.uci.edu/ml.

基于两阶段数据增强和双向深度残差TCN的用户负荷曲线分类方法

Power Load Curve Identification Method Based on Two-Phase Data Enhancement and Bi-directional Deep Residual TCN

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 26

相关文章 7

编辑推荐

Metrics

本文评价

[1]	夏云舒, 王勇, 周林, 樊汝森. 基于改进生成对抗网络的虚假数据注入攻击检测方法[J]. 电力建设, 2022, 43(3): 58-65.
[2]	王磊, 刘洋, 李文峰, 张杰, 许立雄, 邢哲铭. 基于用电行为数字特征画像的电力用户两阶段分类方法[J]. 电力建设, 2022, 43(2): 70-80.
[3]	朱庆, 郑红娟, 唐子逸, 韦思雅, 邹子骁, 吴熙. 基于生成对抗网络的综合能源负荷场景生成方法[J]. 电力建设, 2021, 42(12): 1-8.
[4]	唐子卓, 刘洋, 许立雄, 郭久亿. 基于负荷数据频域特征和LSTM网络的类别不平衡负荷典型用电模式提取方法[J]. 电力建设, 2020, 41(8): 17-24.
[5]	乔文俞，李野，刘浩宇，李扬，杨挺. 基于曲线相似与低秩矩阵的缺失电量数据补全方法[J]. 电力建设, 2020, 41(1): 32-38.
[6]	杨懿男，齐林海，王红，苏林萍. 基于生成对抗网络的小样本数据生成技术研究[J]. 电力建设, 2019, 40(5): 71-77.
[7]	徐衍会，张蓝宇，宋歌. 基于核的模糊C均值逐层聚类算法在负荷分类中的应用[J]. 电力建设, 2015, 36(4): 46-51.