基于MDTimeGAN的序列数据生成方法

doi:10.19678/j.issn.1000-3428.0069688

计算机工程 ›› 2024, Vol. 50 ›› Issue (11): 59-69. doi: 10.19678/j.issn.1000-3428.0069688

基于MDTimeGAN的序列数据生成方法

朱春强¹^,², 刘彬³^,*(), 朱莉³

1. 西安交通大学计算机科学与技术学院, 陕西西安 710049
2. 国网陕西省电力有限公司培训中心, 陕西西安 710032
3. 西安科技大学计算机科学与技术学院, 陕西西安 710054

收稿日期:2024-04-02 出版日期:2024-11-15 发布日期:2024-11-01
通讯作者: 刘彬
基金资助:
国网陕西省电力有限公司科技项目(5226PX240003); 国网陕西省电力有限公司数字化项目(B326PX230001); 国网陕西省电力有限公司数字化项目(B326PX23000)

Sequence Data Generation Method Based on MDTimeGAN

ZHU Chunqiang¹^,², LIU Bin³^,*(), ZHU Li³

1. School of Computer Science and Technology, Xi'an Jiaotong University, Xi'an 710049, Shaanxi, China
2. State Grid Shaanxi Electric Power Company Limited Training Center, Xi'an 710032, Shaanxi, China
3. School of Computer Science and Technology, Xi 'an University of Science and Technology, Xi'an 710054, Shaanxi, China

Received:2024-04-02 Online:2024-11-15 Published:2024-11-01
Contact: LIU Bin

摘要/Abstract

摘要：

非侵入式负荷分解是能源管理领域的一个热门研究课题, 其在各种工业和商业场景中都得到广泛应用。针对负荷分解数据集中存在的样本不平衡问题, 提出一种基于多判别器时间序列生成对抗网络(MDTimeGAN)的序列数据生成方法。通过对原始序列提取时域、频域、时频域以及自相关特征, 并在TimeGAN模型基础上采用4种不同的判别器对时间序列的多维度特征进行判别, 从而提高对原始数据的判别能力, 提升数据质量。在3种公开数据集上进行横向和纵向对比实验, 结果表明, 与对比模型相比, MDTimeGAN模型生成的数据能够更好地覆盖原始数据的分布, 在数据分布方面保持良好的性能, 生成数据符合时间序列数据的特点。

关键词: 非侵入式负荷分解, 时间序列生成对抗网络, 时间序列生成, KS检验, Wassertein距离

Abstract:

Non-invasive load decomposition is a popular research topic in the field of energy management, being widely employed in various industrial and commercial scenarios. To address the problem of sample imbalance in load decomposition datasets, the present study proposes a method for generating sequence data based on Multi-Discriminator Time series Generative Adversarial Network(MDTimeGAN). By extracting the time domain, frequency domain, time-frequency domain, and autocorrelation features of the original sequence and using four different discriminators based on the TimeGAN model to discriminate multidimensional features of the time series, the discrimination ability of the original data and the data quality are improved simultaneously. Horizontal and vertical comparison experiments on three public datasets show that, compared with the comparative models, the data generated by the proposed model more effectively cover the distribution of the original data, maintain good performance in data distribution, and meet the characteristics of time-series data.

Key words: non-invasive load decomposition, Time-series Generative Adversarial Network(TimeGAN), time series generation, KS test, Wassertein distance

朱春强, 刘彬, 朱莉. 基于MDTimeGAN的序列数据生成方法[J]. 计算机工程, 2024, 50(11): 59-69.

ZHU Chunqiang, LIU Bin, ZHU Li. Sequence Data Generation Method Based on MDTimeGAN[J]. Computer Engineering, 2024, 50(11): 59-69.

https://www.ecice06.com/CN/Y2024/V50/I11/59

图/表 18

图1 MDTimeGAN模型结构

Fig.1 MDTimeGAN model structure

图2 数据处理流程

Fig.2 Data processing flow

图3 电力供需平衡数据集纵向对比结果(PCA处理结果)

Fig.3 Longitudinal comparison results of power supply and demand balance dataset(PCA processing results)

图4 电力供需平衡数据集纵向对比结果(t-SNE处理结果)

Fig.4 Longitudinal comparison results of power supply and demand balance dataset(t-SNE processing results)

图5 电力供需平衡数据集横向对比结果(PCA处理结果)

Fig.5 Horizontal comparison results of power supply and demand balance dataset(PCA processing results)

图6 电力供需平衡数据集横向对比结果(t-SNE处理结果)

Fig.6 Horizontal comparison results of power supply and demand balance dataset(t-SNE processing results)

参考文献 27

1	李彬, 李悦欣, 林驿轩, 等. 考虑负荷动态响应特性的居民小微负荷优化调度策略. 供用电, 2024, 41(3): 61- 68.
	LI B, LI Y X, LIN Y X, et al. Optimal scheduling strategy for residential small micro-loads considering dynamic response characteristics of loads. Distribution & Utilization, 2024, 41(3): 61- 68.
2	辛保安. 新型电力系统构建方法论研究. 新型电力系统, 2023, 1(1): 1- 18.
	XIN B A. Research on the methodology of constructing new power systems. New Power System, 2023, 1(1): 1- 18.
3	鲁晓林. 电力系统智能化技术的应用与发展趋势. 光源与照明, 2023,(12): 213- 215.
	LU X L. Application and development trend of intelligent technology of power system. Light Source and Illumination, 2023,(12): 213- 215.
4	王以良, 周鹏, 叶卫, 等. 基于金字塔网络的非侵入式负荷辨识及其隐私保护. 计算机工程, 2024, 50(5): 182- 189. doi: 10.19678/j.issn.1000-3428.0067381
	WANG Y L, ZHOU P, YE W, et al. Pyramid network based non-intrusive load monitoring and its privacy-preserving scheme. Computer Engineering, 2024, 50(5): 182- 189. doi: 10.19678/j.issn.1000-3428.0067381
5	XIA M, WANG Z W, LU M, et al. MFAGCN: a new framework for identifying power grid branch parameters. Electric Power Systems Research, 2022, 207, 107855. doi: 10.1016/j.epsr.2022.107855
6	BASSAMZADEH N, GHANEM R. Multiscale stochastic prediction of electricity demand in smart grids using Bayesian networks. Applied Energy, 2017, 193, 369- 380. doi: 10.1016/j.apenergy.2017.01.017
7	MATEJICEK L. Energy outlook: spatial and temporal mapping of energy sources using GIS[M]//MATEJICEK L. Assessment of energy sources using GIS. Berlin, Germany: Springer, 2017: 61-109.
8	张石清, 王伟, 钱亚冠, 等. 面向深度学习的非侵入式负荷监测研究进展. 计算机系统应用, 2023, 32(3): 25- 47.
	ZHANG S Q, WANG W, QIAN Y G, et al. Deep learning-based non-intrusive load monitoring: recent advances and perspectives. Computer Systems & Applications, 2023, 32(3): 25- 47.
9	姚珺. 基于GAN的多变量时间序列异常检测方法. 安庆师范大学学报(自然科学版), 2023, 29(2): 68- 74.
	YAO J. Multivariate time series anomaly detection via generative adversarial networks. Journal of Anqing Normal University(Natural Science Edition), 2023, 29(2): 68- 74.
10	MOGREN O. C-RNN-GAN: continuous recurrent neural networks with adversarial training[EB/OL]. [2023-05-10]. https://arxiv.org/pdf/1611.09904.
11	van HOUDT G, MOSQUERA C, NÁPOLES G. A review on the long short-term memory model. Artificial Intelligence Review, 2020, 53(8): 5929- 5955.
12	ESTEBAN C, HYLAND S L, RÄTSCH G. Real-valued(medical) time series generation with recurrent conditional GANs[EB/OL]. [2023-05-10]. https://arxiv.org/pdf/1706.02633.
13	YOON J, JARRETT D, van der SCHAAR M. Time-series generative adversarial networks. Advances in Neural Information Processing Systems, 2019, 32(4): 186- 191.
14	DENG Y, LU L, APONTE L. Deep transfer learning and data augmentation improve glucose levels prediction in type 2 diabetes patients. NPJ Digital Medicine, 2021, 4(1): 109.
15	苏鹏, 朱晓荣, 朱洪波. 基于TimeGAN和多臂老虎机的WMN接入模式选择方法. 无线电通信技术, 2023, 49(3): 424- 431.
	SU P, ZHU X R, ZHU H B. Access mode selection method for wireless mesh networks based on TimeGAN and multi-armed bandits. Radio Communication Technology, 2023, 49(3): 424- 431.
16	丁国辉, 刘宇琪, 王言开, 等. 基于翻转网络的低相关性序列数据预测研究. 计算机工程, 2024, 50(2): 78- 90. doi: 10.19678/j.issn.1000-3428.0067027
	DING G H, LIU Y Q, WANG Y K, et al. Research on low-correlation sequence data prediction based on flip network. Computer Engineering, 2024, 50(2): 78- 90. doi: 10.19678/j.issn.1000-3428.0067027
17	陆彦辉, 柳寒, 李航, 等. 基于多鉴别器生成对抗网络的时间序列生成模型. 通信学报, 2022, 43(10): 167- 176.
	LU Y H, LIU H, LI H, et al. Time series generation model based on multi-discriminator generative adversarial network. Journal on Communications, 2022, 43(10): 167- 176.
18	何银银, 胡静, 陈志泊. 融合门控变换机制和GAN的低光照图像增强方法. 计算机工程, 2024, 50(2): 247- 255. doi: 10.19678/j.issn.1000-3428.0067252
	HE Y Y, HU J, CHEN Z B. Low-light image enhancement method combining gated transformation mechanism and GAN. Computer Engineering, 2024, 50(2): 247- 255. doi: 10.19678/j.issn.1000-3428.0067252
19	孙本亮. 基于深度神经网络的负荷监测方法研究[D]. 天津: 河北工业大学, 2021.
	SUN B L. Research on load monitoring method based on deep neural network[D]. Tianjin: Hebei University of Technology, 2021. (in Chinese)
20	刘睿迪. 基于数据增强和深度学习的非侵入式负荷分解方法[D]. 杭州: 浙江大学, 2021.
	LIU R D. Non-intrusive load decomposition based on data augmentation and deep learning[D]. Hangzhou: Zhejiang University, 2021. (in Chinese)
21	崔黎丽, 周云海, 石基辰, 等. 基于爬坡方向状态划分的MCMC风电功率序列建模方法. 现代电子技术, 2024, 47(8): 113- 120.
	CUI L L, ZHOU Y H, SHI J C, et al. Method of MCMC wind power sequence modeling based on climbing direction state division. Modern Electronic Technology, 2024, 47(8): 113- 120.
22	叶宝林, 戴本岙, 张鸣剑, 等. 基于图卷积网络的交通流预测方法综述. 南京信息工程大学学报(自然科学版), 2024, 16(3): 291- 310.
	YE B L, DAI B A, ZHANG M J, et al. A survey of traffic flow prediction methods based on graph convolutional networks. Journal of Nanjing University of Information Science and Technology(Natural Science Edition), 2024, 16(3): 291- 310.
23	陈浩, 杨俊安, 刘辉. 基于深度残差适配网络的通信辐射源个体识别. 系统工程与电子技术, 2021, 43(3): 603- 609.
	CHEN H, YANG J A, LIU H. Communication transmitter individual identification based on deep residual adaptation network. Systems Engineering and Electronics, 2021, 43(3): 603- 609.
24	MAO X D, LI Q, XIE H R, et al. Least squares generative adversarial networks[C]//Proceedings of the IEEE International Conference on Computer Vision(ICCV). Washington D. C., USA: IEEE Press, 2017: 2794-2802.
25	DONG J Y, YIN R Y, SUN X, et al. Inpainting of remote sensing SST images with deep convolutional generative adversarial network. IEEE Geoscience and Remote Sensing Letters, 2019, 16(2): 173- 177.
26	ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein generative adversarial networks[C]//Proceedings of International Conference on Machine Learning. [S. l. ]: PMLR, 2017: 214-223.
27	ZHAO J B, MATHIEU M, LECUN Y. Energy-based generative adversarialnetwork[EB/OL].[2023-05-10]. https://arxiv.org/pdf/1702.01691.

选择文件类型/文献管理软件名称

选择包含的内容

基于MDTimeGAN的序列数据生成方法

Sequence Data Generation Method Based on MDTimeGAN

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 18

参考文献 27

相关文章 1

编辑推荐

Metrics

本文评价

模态框（Modal）标题

选择文件类型/文献管理软件名称

选择包含的内容

基于MDTimeGAN的序列数据生成方法

Sequence Data Generation Method Based on MDTimeGAN

RichHTML

PDF

可视化

被引次数

摘要/Abstract

引用本文

使用本文

图/表 18

参考文献 27

相关文章 1

编辑推荐

Metrics

本文评价