基于多维数据图像化的非侵入式负荷识别  

Non-intrusive load recognition based on multi-dimensional data visualization

作　　者：李成华[1,2] 邹堰辉 江小平[1,2] 石鸿凌 LI Chenghua;ZOU Yanhui;JIANG Xiaoping;SHI Hongling(College of Electronics and Information Engineering,South-Central Minzu University,Wuhan 430074,China;Hubei Province Key Laboratory of Intelligent Wireless Communication,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学电子信息工程学院,湖北武汉430074 [2]中南民族大学智能无线通信湖北省重点实验室,湖北武汉430074

出　　处：《华中科技大学学报（自然科学版）》2023年第5期119-124,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家重点研发计划资助项目(2020YFC1522600);中央高校攻关计划专项资金资助项目(CZT20002).

摘　　要：针对已有负荷识别方法存在选取的负荷印记冗余度大及无法直接反映负荷功率信息的不足,提出一种多维数据图像化的非侵入式负荷识别方法.首先将负荷的电流波形、瞬时功率波形和电压-无功电流轨迹三个维度的负荷印记转换成灰度图像;然后将其分别加载到图像的红绿蓝通道上,得到带有功率信息的真彩色图像;最后通过简化的二维卷积神经网络进行负荷识别.实验结果表明:本方法能够提升图像的信息密度,使得所采用的人工智能网络在计算量和参数量都降低的情况下仍能在图像中找到最具有辨识力的区域进行高效的负荷识别;在PLAID(即插即用设备标识数据集)和WHITED(全球家庭和工业瞬态能量数据集)上分别达到了98.78%和99.50%的识别准确率.To address the limitations of the selected load signature used by existing load recognition methods,which had large redundancy and could not directly reflect the load power information,a non-intrusive load recognition method using multidimensional data visualization was proposed.First,three load signatures,i.e.,the load current waveform,the instantaneous power waveform and the voltage-nonactive current trajectory were converted into gray-scale images,respectively.Then,true-color images with power information were obtained by loading the gray-scale images into three channels,i.e.,red,green and blue.Finally,a simplified two-dimensional convolution neural network for load recognition was constructed.Experimental results show that the proposed method can improve the information density of the image,and the most recognizable area can be still found in the image by using the artificial intelligence network whose computation and parameters were both reduced.The recognition accuracy reaches to 98.78%and 99.50%on PLAID(plug-level appliance identification dataset)and WHITED(worldwide household and industry transient energy dataset),respectively.

关 键 词：非侵入式负荷识别 数据图像化 负荷印记 人工智能网络 数据处理 

分 类 号：TP319.4[自动化与计算机技术—计算机软件与理论]