基于谱图理论的居民用户非侵入式负荷分解  

Non-Intrusive Residential Load Monitor Based on Spectral Graph Theory

作　　者：彭显刚[1] 郑凯[1] 林哲昊 朱俊超[1] 李壮茂 PENG Xiangang;ZHENG Kai;LIN Zhehao;ZHU Junchao;LI Zhuangmao(School of Automation,Guangdong University of Technology,Guangzhou 510006,Guangdong Province,China)

机构地区：[1]广东工业大学自动化学院

出　　处：《电网技术》2018年第8期2674-2680,共7页Power System Technology

摘　　要：针对现有居民用户非侵入式负荷分解需要高频采集数据或大量训练样本的问题,提出了一种基于谱图理论的非侵入式负荷分解方法。首先,以用户总负荷采样信号的相邻采样点差值来建立图结构,同时通过对用电设备信号采样点差值的分类来定义用电设备的图信号;然后,通过图拉普拉斯变换得到的图信号全局平滑度函数来实现用电设备图信号未知部分的重构。在用电设备时序信号的重构过程中采用了模糊规整方法来解决采样信号平滑性所导致的重构图信号数值的非标准化问题;对重构的用电设备图信号中相邻非零值间所对应的同时段负荷时序信号值赋以该用电设备的对应状态数据,从而实现用电设备时序信号的重构。最后采用AMPds数据集进行了仿真实验,结果表明所提方法有效且实用,能够在较低采样频率和先验信息较少的条件下实现较高精度的负荷分解。A non-intrusive load monitor（NILM） method based on spectral graph theory（SGT） is proposed to solve the problem of requirement of high frequency acquisition data and large number of training samples in NILM. Firstly, a graph structure is established using difference of adjacent sample signals from total load signals and a graph signal is defined by classifying difference of adjacent samples from appliance load signals. Secondly, unknown graph signals from appliance load signals are reconfigured based on minimum total smoothness function（TSF） obtained with graph Laplace transform. Fuzzy regularization method is used to figure out normalization problem in signal smoothing process. Interval time points between adjacent non-zero values in the reconfigured graph signals are assigned to corresponding state data of the appliance, then time series signals of each appliance are reconfigured. Finally, AMPds data set is used to carry out simulation. Result shows that the proposed method is effective and feasible. This method can achieve higher accuracy load decomposition under the condition of lower sampling frequency and less prior information.

关 键 词：非侵入式负荷分解 谱图理论 图拉普拉斯变换 信号恢复 

分 类 号：TM714[电气工程—电力系统及自动化]