基于堆叠稀疏自编码器与GAN的线损分层定位
Layered line loss localization based on stacked sparse autoencoder and GAN
杜月 1王慧琴 1余兆媛 1钱亚林 1魏敏俊1
作者信息
- 1. 国网安徽省电力有限公司安庆供电公司,安徽安庆 246003
- 折叠
摘要
线损数据中存在噪声数据,维度与线损数据一致,导致线损定位结果不精准.为此,提出基于堆叠稀疏自编码器与GAN的线损分层定位方法.构建线损分层定位GAN结构,判别假数据和真数据,获取线损分层数据.依据分层采集结果,计算分层供入、供出电量和统计线损,以此作为分层存在异常线损的依据.基于堆叠稀疏自编码器的定位原理,通过在代价函数中增加散度,引导输出结果稀疏.根据确定的稀疏编码所在空间,借助SVM分类核函数,定位线损所在层次.由实验结果可知,所研究方法统计的四种线损变化范围分别是3.0～4.0 kW·h、1.0～3.0 kW·h、0.4～0.7 kW·h、4.2～4.8 kW·h,对应的窃电位置分别为表箱1层、表箱2层、表箱3层、表箱4层,具有精准的定位效果.
Abstract
Due to the presence of noise data in the line loss data,the dimensionality is consistent with the line loss data,resulting in inaccurate line loss localization results.Therefore,a hierarchical line loss localization method based on stacked sparse autoencoder and GAN is proposed.Construct a hierarchical positioning GAN structure for line loss,distinguish between false and true data,and obtain hierarchical data for line loss.Based on the layered collection results,calculate the layered incoming and outgoing electricity quantities and calculate the line losses,as a basis for the existence of abnormal line losses in the layered data.Based on the positioning principle of stacked sparse autoencoder,increasing divergence in the cost function leads to sparse output results.Based on the determined sparse encoding space,the SVM classification kernel function is used to locate the level of line loss.According to the experimental results,the four types of line loss variation ranges calculated by the research method are 3.0～4.0 kW·h,1.0～3.0 kW·h,0.4～0.7 kW·h,4.2～4.8 kW·h,respectively.The corresponding stealing positions are meter box 1 layer,meter box 2 layer,meter box 3 layer,and meter box 4 layer,which have precise positioning effects.
关键词
堆叠稀疏自编码器/生成对抗网络/线损分层定位/SVM分类/稀疏编码Key words
stacked sparse autoencoder/Generative Adversarial Networks/layered line loss localization/SVM classification/sparse coding引用本文复制引用
出版年
2025
NETL
NSTL
万方数据