摘要
为进一步提升信息化电网性能,提出一种基于长短期记忆网络(LSTM)与压缩感知(CS)实现电物理量轻型化方法。将时间序列变化的电物理量采样数据作为LSTM模型的输入量,稳定结果作为输出量,通过时间反向传播(BPTT)算法训练LSTM模型参数,训练后的模型能充分利用循环架构的特征进行模式识别,并根据LSTM模式识别结果,将信号选择原子库进行CS,来确定测量采样频率。实例结果表明,LSTM+CS方法比常用传统方法采样频率低,传输参数少,可大幅节省存储容量和减轻网络流量。
To further improve the performance of the informationalized power grid,a long-short-term memory(LSTM)network and compressed sensing(CS)method are put forward to realize the lightening of electric physical quantity.The sampled data of time series changes in electrical physical quantities are used as the input of the LSTM model,and the stable results are used as output.LSTM model parameters are trained by back propagation trough time(BPTT)algorithm.After training,the model can make full use of the characteristics of the cyclic structure for pattern recognition.According to LSTM pattern recognition results,the signal selects atom library for CS to determine the measurement sampling frequency.The simulation results show that the LSTM+CS method has lower sampling frequency and less transmission parameters than traditional methods,so it can significantly save storage capacity and reduce the network traffic.
作者
周学斌
李晓明
李雷
甘凌霞
ZHOU Xuebin;LI Xiaoming;LI Lei;GAN Lingxia(School of Electrical Engineering,Wuhan University,Wuhan 430072,China;Jiujiang Power Supply Branch of Jiangxi Electric Power Co.Ltd.,Jiujiang 332000,China)
出处
《电力系统自动化》
EI
CSCD
北大核心
2019年第1期102-109,241,共9页
Automation of Electric Power Systems
基金
国家自然科学基金资助项目(51277134)~~
关键词
电力系统
轻型化
深度学习
长短期记忆网络
压缩感知
power systems
lightening
deep learning
long-short-term memory (LSTM)network
compressed sensing (CS)
