肖琦, 陆俊, 孙霄羽, 龚钢军, 王振宇. 基于插值优化的联邦学习异常用电辨识研究[J]. 电力信息与通信技术, 2025, 1(1): 1-9. DOI: 10.16543/j.2095-641x.electric.power.ict.2025.01.01
引用本文: 肖琦, 陆俊, 孙霄羽, 龚钢军, 王振宇. 基于插值优化的联邦学习异常用电辨识研究[J]. 电力信息与通信技术, 2025, 1(1): 1-9. DOI: 10.16543/j.2095-641x.electric.power.ict.2025.01.01
XIAO Qi, LU Jun, SUN Xiaoyu, GONG Gangjun, WANG Zhenyu. Research on Abnormal Power Consumption Identification of Federal Learning Based on Interpolation Optimization[J]. Electric Power Information and Communication Technology, 2025, 1(1): 1-9. DOI: 10.16543/j.2095-641x.electric.power.ict.2025.01.01
Citation: XIAO Qi, LU Jun, SUN Xiaoyu, GONG Gangjun, WANG Zhenyu. Research on Abnormal Power Consumption Identification of Federal Learning Based on Interpolation Optimization[J]. Electric Power Information and Communication Technology, 2025, 1(1): 1-9. DOI: 10.16543/j.2095-641x.electric.power.ict.2025.01.01

基于插值优化的联邦学习异常用电辨识研究

Research on Abnormal Power Consumption Identification of Federal Learning Based on Interpolation Optimization

  • 摘要: 爆发式增长的用电用户数据为数据驱动的异常用电辨识奠定了基础,然而不同电力公司间的“数据壁垒”以及用电数据采集频率不高的问题会影响此类方法的性能。为了解决异常用电行为分布式辨识中数据异构和采集频次不高影响辨识性能的问题,文章提出一种基于插值优化的联邦学习异常用电辨识方法。首先,对待辨识异常用电数据集进行线性插值与快速傅里叶变换处理。其次,构建基于长短期记忆机制的分布式联邦学习训练系统。最后,利用构建的训练模型,实现训练系统异构下的分布式异常用电行为辨识。所提方法在某电网大区的真实数据上进行实验,实验结果表明,方法能够实现对数据集的数据增强,在分布式训练与训练数据非独立同分布条件下对用电用户的异常用电行为进行有效识别,有效辅助窃电行为稽查,提升电网运维效率。

     

    Abstract: The explosive growth of electricity user data has laid the foundation for data-driven identification of abnormal electricity consumption. However, the ' data barriers ' between different power companies and the low frequency of electricity data collection will affect the performance of such methods. In order to solve the problem that data heterogeneity and low acquisition frequency affect the identification performance in the distributed identification of abnormal power consumption behavior, this paper proposes a federal learning abnormal power consumption identification method based on interpolation optimization. Firstly, linear interpolation and FFT processing are performed on the data set of abnormal electricity consumption to be identified. Secondly, a distributed federated learning and training system based on LSTM mechanism is constructed. Finally, using the constructed training model, the distributed abnormal power consumption behavior identification under the heterogeneous training system is realized. The proposed method is tested on the real data of a large area of a power grid. The experimental results show that the method can realize the data enhancement of the data set, and effectively identify the abnormal electricity consumption behavior of electricity users under the condition that the distributed training and training data are not independent and identically distributed, effectively assist the electricity stealing behavior audit, and improve the operation and maintenance efficiency of the power grid.

     

/

返回文章
返回