基于LSTM的三相电压不平衡监测  

Monitoring Method for Three-phase Voltage Imbalance Based on LSTM

作　　者：王海燕 杨照 WANG Haiyan;YANG Zhao(China Coal Huajin Group Co.Ltd.,Hejin Shanxi 043300;China University of Mining and Technology,Xuzhou Jiangsu)

机构地区：[1]中煤华晋集团有限公司,山西河津043300 [2]中国矿业大学,江苏徐州221116

出　　处：《中国科技纵横》2023年第9期107-111,共5页China Science & Technology Overview

摘　　要：基于深度学习(DL)的三相电压不平衡度自动特征提取和预测方法由专用的长短期记忆(LSTM)网络组成,这是一种特殊的递归神经网络(RNN)。运用从某电网10kV进线侧采集到的海量三相电压不平衡度RMS序列。结果表明,提出的方法可以根据LSTM中的学习特征对三相电压不平衡度进行预测,在测试数据集上的预测精度为93.40%,开发的网络模型对于特征学习和不平衡度的预测是精确的。与常规的机器学习方法不同,当有大量测量数据可用时,本文所提出的方法能够学习倾角特征,而无需过渡事件分割选择阈值以及使用专家规则或人类专家知识。这为利用深度学习技术进行电能质量数据分析和分类开辟了新的可能性。An automatic feature extraction and prediction method of three-phase voltage imbalance based on deep learning(DL)consists of a dedicated long short-term memory(LSTM)network,which is a special kind of recurrent neural network(RNN).This paper uses a large-scale three-phase voltage imbalance RMS sequence collected from the 10kV incoming side of a certain power grid.Our results show that the proposed method can predict the three-phase voltage imbalance based on the learning characteristics in LSTM,and the classification accuracy on the test data set is 93.40%,network model developed in this paper is accurate for feature learning and prediction of imbalance.Unlike conventional machine learning methods,when a large amount of measurement data is available,the method proposed in this paper can learn the inclination features without the need for transition event segmentation,selection of thresholds,and use of expert rules or human expert knowledge.This opens up new possibilities for deep quality analysis and classification of power quality data.

关 键 词：深度学习 LSTM 三相电压不平衡度预测 电能质量 

分 类 号：TH311.13[机械工程—机械制造及自动化]