基于频谱图与时序成像的非侵入式负荷监测方法

Non-intrusive Load Monitoring Method Based on Spectrum Map and Time-sequence Imaging

针对在多种电器设备同时运行的场景下,当前非侵入式负荷监测方法存在分解困难的问题,本文提出了基于频谱图与时序成像的非侵入式负荷监测方法。首先进行负荷分解,利用频谱图变换原理将多种电器设备的聚合电流转换成频谱图矩阵,并通过词嵌入将频谱图矩阵变换到高维;然后通过k-均值聚类算法得到单个电器设备的频谱图矩阵并反变换为相应的时序电流;其次,进行负荷分类,将负荷分解得到的各类电器设备的时序电流转换为图像进行分类,分类模型为训练完成的深度神经网络模型。最后,利用公开数据集进行实验,结果表明所提方法具有较好的分解和分类效果。

In the scenario of multiple electrical equipment running at the same time,the existing non-intrusive load monitoring(NILM)method has problems such as difficulty in decomposition.Aimed at this problem,an NILM method based on spectrum map and time-sequence imaging is proposed in this paper.First,load decomposition is carried out,the aggregated current of multiple electrical equipment is converted into a spectrum map matrix according to the spec-trum map transformation principle,and the spectrum map matrix is expanded to high dimensions through word embed-ding.Afterwards,the spectrum map matrix of one single electrical equipment is obtained using the k-means clustering algorithm and inversely transformed into the corresponding time-sequence current.Second,load classification is per-formed,in which the time-sequence current of multiple electrical equipment obtained by load decomposition is convert-ed into images for classification and a trained deep neural network model is taken as the classification model.Finally,the results of experiment on a public data set show that the proposed method has satisfying decomposition and classifica-tion effects.