摘要
针对配电系统多时刻量测缺失数据修复因误差累积导致准确率降低的问题,提出了一种基于多步长长短期记忆神经网络(multi-step long-short term memory,MLSTM)和协方差交叉(covariance intersection,CI)融合的配电系统多时刻量测缺失数据修复方法。首先,将配电系统电流、功率等量测量历史数据降维后,构建不同维度的输入向量矩阵和特征标签矩阵作为模型输入,并训练得到多个不同步长的长短期记忆神经网络(long-short term memory,LSTM)量测数据修复模型。在此基础上,利用CI算法对上述不同步长的LSTM修复模型进行融合,得到多时刻量测缺失数据修复模型。算例分析表明,所提方法可以有效抑制多时刻量测数据修复过程中的误差累积,提高多时刻缺失数据的修复准确度。
Focusing on the problem of accuracy reduction due to error accumulation in multi-moment missing measurement data reparation of the power distribution system,this paper proposes a multi-moment missing measurement data repair method based on multi-step long-short term memory(MLSTM)and covariance intersection(CI)for the power distribution system.Firstly,the historical data of current,power,and other quantity measurements of power distribution system are downscaled to construct the input vector matrix and feature label matrix with different dimensions as model inputs and then trained to obtain multiple long-short term memory(LSTM)measurement data repair models with different time steps.On this basis,the CI algorithm is utilized to fuse the above LSTM repair models with different time steps to obtain a multi-moment missing measurement data repair model for the power distribution system.Example analysis shows that the proposed method can effectively suppress the error accumulation in repairing multi-moment measurement data and improve the reparation accuracy of multi-moment missing data.
作者
郭凌旭
王天昊
黄盼
王冬阳
李振斌
GUO Lingxu;WANG Tianhao;HUANG Pan;WANG Dongyang;LI Zhenbin(State Grid Tianjin Electric Power Company,Tianjin 300143,China;Electric Power Research Institute of State Grid Tianjin Electric Power Company,Tianjin 300384,China;Key Laboratory of Smart Grid of Ministry of Education,Tianjin University,Tianjin 300072,China;State Grid Tianjin Electric Power Company Chengnan Power Supply Branch,Tianjin 300202,China)
出处
《山东电力技术》
2025年第8期56-66,共11页
Shandong Electric Power
基金
国网天津市电力公司科技项目(电科-研发2023-51)
国家重点研发计划资助项目(2020YFB0905900)。
关键词
配电系统
量测缺失数据修复
长短期记忆神经网络
协方差交叉
power distribution system
missing measurement data reparation
long-short term memory neural network
covariance intersection
