摘要
针对非侵入负荷分解如何实现未识别设备负荷事件的准确检测的问题,设计云-边协同计算体系,通过挖掘边缘侧采集的负荷数据行为特征,实现有效分解负荷信号。考虑边缘侧嵌入式控制平台计算与存储资源及功耗等指标限制,提出轻量化K均值聚类算法;选取瞬时峰值、瞬时功率变化量及谐波分量为聚类特征在本地实现负荷类型辨识;设计轻量化粒子群优化算法,实现动态聚类算法与暂态功率、谐波分量等特征的结合。本文设计的电能监测算法对REDD共享数据集的识别准确率大于97.79%,实现了对某小区313户居民日用电数据的非侵入负荷监测。
The key problem in non-intrusive load decomposition is how to realize an accurate detection of load events of unidentified equipment in real time.Aimed at this problem,a cooperative computing system on the cloud-edge side is designed,which mines the behavior feature of load data collected on the edge side and hence realizes an effective decomposition of load signal.A lightweight K-means clustering algorithm is proposed considering the constraints of indexes such as the computing and storage resources of embedded controllers on the edge side together with power consumption.The instantaneous peak value,instantaneous power variation and harmonic components are selected as clustering features to realize the identification of load category locally.A lightweight particle swarm optimization algorithm is designed to realize the combination of the dynamic clustering algorithm and features including the transient power and harmonic components.The identification accuracy of the electric energy monitoring algorithm designed in this paper is higher than 97.79%for the REDD shared data set.In addition,the non-invasive load monitoring of daily power consumption data for 313 households in one community is realized.
作者
蔡田田
杨英杰
陈波
邓清唐
CAI Tiantian;YANG Yingjie;CHEN Bo;DENG Qingtang(Digital Grid Research Institute,China Southern Power Grid,Guangzhou 510700,China)
出处
《电力系统及其自动化学报》
CSCD
北大核心
2023年第12期50-58,共9页
Proceedings of the CSU-EPSA
基金
国家重点研发计划资助项目(2020YFB0906000,2020YFB0906002)。
关键词
边缘计算
负荷辨识
群智能
负荷分解
特征聚类
edge computing
load identification
swarm intelligence
load decomposition
feature clustering