基于EMD-E-LSTM模型的气体浓度预测模型设计

Design of gas concentration prediction model based on EMD-E-LSTM model

龙玉江 ¹姜超颖 ²甘润东 ¹吴建蓉³

扫码查看

作者信息

1. 贵州电网有限责任公司信息中心,贵州贵阳 550003
2. 西安电子科技大学,陕西西安 710071
3. 贵州电科院,贵州贵阳 550003
折叠

摘要

为了完成检测、维修等重要工作,针对变压器油中溶解气体浓度变化趋势预测的问题,文中将经验模态分解(Empirical Mode Decomposition,EMD)方法、编码器(Encoder)模块和长短时记忆(Long Short-Term Memory,LSTM)神经网络相结合,提出一种新颖的EMD-E-LSTM网络预测模型来实现油中溶解气体浓度的预测.对110 kV变压器油中溶解C2H6气体浓度预测算例结果表明,相较于E-LSTM预测方法、EMD-LSTM预测方法,所提EMD-E-LSTM网络预测结果的平均绝对百分比误差分别降低了22.23%和5.50%、均方根误差分别下降了18.18%和44.02%,而最大相对误差介于两者之间.所提方法对于其他溶解气体的预测精度也有不同程度的提高,展现出良好的应用前景.

Abstract

For the purpose of predicting the change trend of dissolved gas concentration in transformer oil,and then completing important work such as detection and maintenance.In this paper,the empirical mode decomposition EMD method,Encoder module and the LSTM neural network are combined.A novel EMD-E-LSTM network prediction model is proposed to predict dissolved gas concentration in oil.The prediction results of dissolved C2H6 gas concentration in 110 kV transformer oil show that compared to the E-LSTM prediction method and the EMD-LSTM prediction method,the MAPE of the proposed EMD-E-LSTM network prediction results have decreased by 22.23%and 5.50%,and the root-mean-square error has decreased by 18.18%and 44.02%,while the maximum relative error was between the two.The proposed method can also improve the prediction accuracy of other dissolved gases,showing a good application prospect.

关键词

变压器/经验模态分解/注意力机制/循环神经网络/长短时记忆神经网络

Key words

transformer/EMD/attention mechanism/recurrent neural network/LSTM neural network

引用本文复制引用

出版年

2025

电子设计工程

西安三才科技实业有限公司

电子设计工程

影响因子：0.333

ISSN：1674-6236

段落导航