| 引用本文: | 周勇军,吴元香,董智华,胡誉蓉,肖先勇,张姝.基于模体挖掘与调和函数半监督学习的非侵入式负荷监测[J].电力自动化设备,2022,42(7): |
| ZHOU Yongjun,WU Yuanxiang,DONG Zhihua,HU Yurong,XIAO Xianyong,ZHANG Shu.Non-intrusive load monitoring based on motif mining and harmonic function based semi-supervised learning[J].Electric Power Automation Equipment,2022,42(7): |
|
| 摘要: |
| 针对现有非侵入式负荷监测(NILM)方法成本高昂等问题，提出一种基于模体挖掘与调和函数半监督学习的NILM方法。基于低频采样数据，根据从监测数据得到的功率阶跃量，利用时间序列分析法和模体挖掘法划分设备的运行窗；在设备运行窗中，根据设备特性与统计方法定义设备开启最大值到稳定运行点的斜率、设备稳定运行时的波动幅度2个新的特征量；构建设备运行窗的特征向量，并利用基于调和函数的半监督学习算法对运行窗中的设备类型进行识别。基于参考能量分解数据集，分别从事件匹配和设备识别的角度将模体挖掘和基于调和函数的半监督学习算法与其他NILM方法进行对比，验证了所提方法的准确性和可推广性。 |
| 关键词: 非侵入式负荷监测 时间序列 模体挖掘 调和函数 半监督学习 |
| DOI：10.16081/j.epae.202202013 |
| 分类号:TM73 |
| 基金项目:国家自然科学基金资助项目(52007126)；国家电网公司西藏电力有限公司科技项目(52311020009X) |
|
| Non-intrusive load monitoring based on motif mining and harmonic function based semi-supervised learning |
|
ZHOU Yongjun1,2, WU Yuanxiang3, DONG Zhihua1, HU Yurong2, XIAO Xianyong2, ZHANG Shu2
|
|
1.Lhasa Power Supply Company of State Grid Tibet Electric Power Co.,Ltd.,Lhasa 850010, China;2.College of Electrical Engineering, Sichuan University, Chengdu 610065, China;3.State Grid Tibet Electric Power Co.,Ltd.,Lhasa 850000, China
|
| Abstract: |
| Aiming at the problems such as high cost of existing NILM(Non-Intrusive Load Monitoring) methods, a NILM method based on motif mining and harmonic function based semi-supervised learning is proposed. Based on low-frequency sampling data, according to the power step quantity obtained from the monitoring data, the time series analysis method and motif mining method are used to partition the equipment operation window. In the equipment operation window, two new characteristic quantities of slope from the maximum value when equipment opens to the stable operation point, and the fluctuation amplitude during stable operation of the equipment are defined according to the characteristics of the equipment and statistical methods. The feature vector of equipment operation window is constructed, and the harmonic function based semi-supervised learning algorithm is used to identify the equipment type in the operation window. Based on the reference energy disaggregation data set, the motif mining and harmonic function based semi-supervised learning algorithm are compared with other NILM methods respectively from the perspectives of event matching and equipment identification, and the accuracy and generalization of the proposed method are verified. |
| Key words: non-intrusive load monitoring time series motif mining harmonic function semi-supervised learning |
