基于MDTimeGAN的序列数据生成方法  

Sequence Data Generation Method Based on MDTimeGAN

作　　者：朱春强 刘彬 朱莉[3] ZHU Chunqiang;LIU Bin;ZHU Li(School of Computer Science and Technology,Xi′an Jiaotong University,Xi′an 710049,Shaanxi,China;State Grid Shaanxi Electric Power Company Limited Training Center,Xi′an 710032,Shaanxi,China;School of Computer Science and Technology,Xi′an University of Science and Technology,Xi′an 710054,Shaanxi,China)

机构地区：[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

出　　处：《计算机工程》2024年第11期59-69,共11页Computer Engineering

基　　金：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,B326PX23000)。

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。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.

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 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]