基于深度学习算法的非侵入式工业负荷分解方法  

Non-intrusive load monitoring of industrial load based on deep learning algorithm

作　　者：尹利 于波 马飞 朱生荣 金溢杭 袁志昌[2] 熊春晖 YIN Li;YU Bo;MA Fei;ZHU Sheng-rong;JIN Yi-hang;YUAN Zhi-chang;XIONG Chun-hui(Ulanqab Electric Power Bureau, Inner Mongolia Electric Power (Group) Co., Ltd., Ulanqab 100176, China;Department of Electrical Engineering, Tsinghua University, Beijing 100084, China;Beijing Zhizhong Energy Internet Research Institute Co., Ltd., Beijing 100048, China)

机构地区：[1]内蒙古电力(集团)有限责任公司乌兰察布电业局,内蒙古乌兰察布100176 [2]清华大学电机工程与应用电子技术系,北京100084 [3]北京智中能源互联网研究院有限公司,北京100048

出　　处：《电工电能新技术》2022年第5期61-70,共10页Advanced Technology of Electrical Engineering and Energy

基　　金：内蒙古电力(集团)有限责任公司项目(WD-ZXZB-2020-WZ0203-1853)。

摘　　要：对工业用户进行负荷感知、分析和管理是推进智能电网建设的重要一步。非侵入式负荷分解需要根据用户总负荷来推断识别用户用电设备的负荷并提高电力用户的隐私安全。以往负荷建模和分解的对象主要为家用电器负荷,针对工业用户的负荷分解具有负荷特征较少和设备状态难以建模的问题。本文提出一种对设备差分负荷频谱曲线聚类的方法建模工业设备的负荷状态,并将负荷状态作为分解目标。然后,以具有时间依赖关系的双向长短期记忆神经网络为主干构建了一个由总负荷序列到各设备负荷状态序列的分解模型。最后,基于乌兰察布地区某工厂的月度负荷数据开展案例研究,使用本文方法得到的四台设备负荷状态平均准确率、精确率、召回率、F1_score值均超过90%,取得了良好的分解效果。Load sensing,analysis and management for industrial users is an important step to promote the construction of smart grid.Non-intrusive load monitoring needs to infer and identify the style and state of load of users’electrical equipment according to the total load of users,and improve the privacy security of users.In the past,the load modeling and disaggregation mainly focus on household appliance load.In order to solve the problem that characteristics of industrial load are less and the load states are difficult to model,a method based on spectral curve clustering of equipment differential load is proposed in this paper to model the load of industrial equipment,which can quickly obtain the load states of industrial equipment.Furthermore,the target of load disaggregation is replaced by the state of load instead of primal load.Then,the bi-directional long short-term memory neural network is used to learn the load disaggregation with the total load known.In the case study,based on the equipment load of a factory in Ulanqab area,the load state is obtained by load modeling,and the test accuracy is over 90%after learning by using long short-term memory neural network.

关 键 词：负荷分解 工业负荷 非侵入式 聚类 深度学习 

分 类 号：TM714[电气工程—电力系统及自动化]