基于LSTM和随机矩阵理论的配电网线损异常诊断方法
Abnormal Diagnosis Method of Distribution Network Line Losses Based on the LSTM and Random Matrix Theory
赵卓 1王晓东 1关景林 1张婧1
作者信息
- 1. 国网辽宁省电力有限公司鞍山供电公司,辽宁鞍山 114000
- 折叠
摘要
配电网线损精细化管理是电力系统运行、规划的重要任务,其数据来源于多种电力业务系统,呈现出层次多、类型多、体量大的特性,数据缺失、线损异常、数据难以溯源等问题是配电网线损精细化管理亟需解决的难题.提出一种基于长短期记忆网络(long short-term memory network,LSTM)和随机矩阵理论(random matrix theory,RMT)的线损异常诊断方法.在线损数据缺失的情况下,基于长短期记忆网络对线损数据进行预测,补全缺失的时空线损数据.利用随机矩阵理论的平均谱半径和特征根密度值作为量化指标,建立线损异常诊断模型,对线损数据进行特征分析,从而实现对配电网线损异常的诊断.利用电网某配电台区的实际线损数据进行验证,结果表明,该方法在数据缺失或异常时,仍能够保证其在配电网线损异常诊断方面的有效性和准确性.
Abstract
The refined management of distribution network line losses is an important task for operation and planning of the power system.The data of the line loss management system comes from various electric power business systems,presenting characteristics of multiple levels,types,and large volumes.The urgent issue facing the refined management of distribution network line losses is how to address problems such as missing data,abnormal line loss,and difficulty in tracing data.Therefore,this paper proposes a line loss anomaly diagnosis method based on Long short-term memory network(LSTM)and random matrix theory(RMT).Firstly,in the case of missing data,the line loss data is predicted based on long-term and short-term memory networks to complete the missing spatiotemporal line loss data.Second,using the mean spectral radius and characteristic root density values of random matrix theory as quantitative indicators,a line loss anomaly diagnosis model is established,and feature analysis is performed on the line loss data to achieve the diagnosis of distribution network line loss anomalies.Finally,the effectiveness and accuracy of the proposed method in diagnosing abnormal line losses in distribution networks are verified using the actual line loss data from a certain distribution station area of the power grid.
关键词
线损/长短期记忆网络/随机矩阵理论/平均谱半径/特征根密度值Key words
line loss/LSTM/RMT/mean spectral radius/characteristic root density value引用本文复制引用
基金项目
国家电网有限公司科技项目(SGLNJX00YJJS2100620)
出版年
2024
NETL
NSTL
万方数据