| 引用本文: | 张 帅,程志友,田 甜,徐正林,杨 猛.基于马尔可夫转移场和轻量级网络的非侵入式负荷识别[J].电力系统保护与控制,2024,52(17):51-61.[点击复制] |
| ZHANG Shuai,CHENG Zhiyou,TIAN Tian,XU Zhenglin,YANG Meng.Non-intrusive load identification based on the Markov transition field and a lightweight network[J].Power System Protection and Control,2024,52(17):51-61[点击复制] |
|
| 摘要: |
| 负荷识别是非侵入式负荷监测(non-intrusive load monitoring, NILM)的关键一步。针对目前识别方法存在网络参数量大、识别率低的局限性，提出了一种基于马尔可夫转移场((Markov transition field, MTF)和轻量级网络的非侵入式负荷识别方法。首先，利用归一化后的电压电流计算马尔可夫状态转移矩阵，在时域上排列每个状态转移概率构建MTF。其次，对MTF降采样以适应神经网络的学习，利用伪彩色编码技术得到RGB彩色图像。最后，在轻量级网络ShuffleNetV2中加入SimAM无参注意力模块作为特征提取网络，以较少的参数量实现负荷分类识别。使用公共数据集PLAID和WHITED对所提方法进行实验，结果表明，SimAM-ShuffleNetV2在两个数据集的识别准确率分别达到了98.99%和99.22%，参数量分别为0.37 M和0.41 M，比现有的方法具有更高的识别准确率和更少的参数量，验证了所提方法的有效性和优越性。 |
| 关键词: 非侵入式负荷识别 数据图像化 马尔可夫转移场 SimAM无参注意力 轻量级网络 |
| DOI：10.19783/j.cnki.pspc.231443 |
| 投稿时间：2023-11-10修订日期：2024-01-02 |
| 基金项目:国家自然科学基金项目资助(61672032)；安徽省自然科学基金项目资助(2108085QE237) |
|
| Non-intrusive load identification based on the Markov transition field and a lightweight network |
| ZHANG Shuai1,CHENG Zhiyou1,2,TIAN Tian1,XU Zhenglin1,YANG Meng3 |
| (1. School of Internet, Anhui University, Hefei 230039, China; 2. Power Quality Engineering Research Center
(Anhui University), Ministry of Education, Hefei 230601, China; 3. School of Electronics and
Information Engineering, Anhui University, Hefei 230601, China) |
| Abstract: |
| Load identification is a key step in non-intrusive load monitoring (NILM). There are limitations caused by a large number of network parameters and a low identification rate of current identification methods. Thus a non-intrusive load identification method based on a Markov transition field (MTF) and a lightweight network is proposed. First, the Markov state transition matrix is calculated using the normalized voltage and current, and the MTF is constructed by arranging each state transition probability in the time domain. Second, the MTF is downsampled to facilitate the learning of the neural network, and an RGB color image is obtained using a pseudo-color coding technique. Finally, a SimAM parameter-free attention module is added to the lightweight network ShuffleNetV2 as a feature extraction network to achieve load classification identification with fewer parameters. Experiments on the proposed method using the public datasets PLAID and WHITED show that the identification accuracy of SimAM-ShuffleNetV2 achieves 98.99% and 99.22% respectively in the two datasets, with the number of parameters of 0.37 M and 0.41 M, respectively. This shows higher identification accuracy and fewer parameters than existing methods, verifying the validity and superiority of the proposed method. |
| Key words: non-intrusive load identification data visualization Markov transition field SimAM parameter-free attention lightweight network |
