基于改进K-means++和LSTM算法的居民负荷远程分解方法  

Remote Decomposition Method for Residential Load Based on Improved K-means++ and LSTM algorithm

作　　者：廖贺 喻伟[1] 熊政[1] 豆龙龙 周惯衡 LIAO He;YU Wei;XIONG Zheng;DOU Longlong;ZHOU Guanheng(Jiangsu Frontier Electric Power Technology Co.,Ltd.,Nanjing 211106,China)

机构地区：[1]江苏方天电力技术有限公司,江苏南京211106

出　　处：《湖南电力》2025年第1期93-99,共7页Hunan Electric Power

摘　　要：针对低压居民用户数量庞大,额外安装监测设备或升级现有监测设备成本高昂的问题,基于高级量测体系的大规模分钟级采集数据,提出一种改进K-means++和长短时记忆算法的居民负荷远程分解方法。首先,基于滑动窗的双边累计和算法监测更加精准、高效、实时地捕捉数据的变化。其次,采用改进的K-means++算法找到具有代表性的负荷进行负荷识别且保证运算速率。最后利用长短时记忆算法,捕捉随着时间发生规律性变化的数据来完成负荷分解。通过在1 min的低采样频率下采集的居民日常负荷数据,充分验证了算法的适用性。In response to the large number of low-voltage residential users and the high cost of installing additional monitoring equipment or upgrading existing monitoring equipment,an improved K-means++and long short-term memory algorithm based on advanced measurement system for large-scale minute level data collection is proposed for remote load decomposition of residents.Firstly,in order to capture changes in data more accurately and efficiently,and in real-time a sliding window based bilateral accumulation and algorithm monitoring is designed.Secondly,in order to identify representative loads and ensure computation speed,an improved K-means++algorithm is adopted.Finally,using long short-term memory algorithms,data that undergoes regular changes over time are captured to complete load decomposition.At a low sampling frequency of 1 minute,the applicability of the proposed algorithm is fully validated through the collection of daily load data from residents.

关 键 词：高级量测体系 K-Means++算法 LSTM算法 居民负荷 远程分解 

分 类 号：TM714[电气工程—电力系统及自动化]