引用本文:
【打印本页】   【HTML】 【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览 1841次   下载 64 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于BAMCNN-GRU混合模型的辅助动力装置排气温度预测方法
何永勃,曹祝兵,于洁
中国民航大学 电子信息与自动化学院,天津 300300
摘要:
对飞机辅助动力装置(Auxiliary Power Unit,APU)排气温度(Exhaust Gas Temperature,EGT)的准确预测可为APU健康管理提供重要信息。传统方法在长周期预测中精度较低。提出一种基于特征与时序的双侧注意力机制(Bilateral Attention Mechanism,BAM)和卷积神经网络(Convolutional Neural Network,CNN)-门控循环单元(Gated Recurrent Unit,GRU)的混合模型,选取5个与排气温度关联度较高的特征参数对EGT进行多变量预测。引入BAM可自动量化输入变量与EGT的关联度,并加强历史关键信息对预测输出的表达;引入CNN可提取反映EGT非平稳动态变化的高维特征。实验结果表明:所提出的混合模型在单步与多步的长时间序列和多变量输入EGT预测均取得很好的效果。相比于BAM-GRU模型、CNN-GRU模型、GRU模型、长短期记忆(Long Short-Term Memory,LSTM)模型、支持向量机(Support Vector Machine,SVM)模型和反向传播(Back Propagation,BP)模型,混合模型的预测精度有较大程度提高。
关键词:  辅助动力装置  排气温度  门控循环单元  特征注意力机制  时序注意力机制  卷积神经网络
DOI:10.13675/j.cnki.tjjs.210351
分类号:V231.1
基金项目:民航科技项目基金(MHRD20150220)。
Prediction Method of Auxiliary Power Unit Exhaust Temperature Based on BAM and CNN-GRU Mixed Model
HE Yong-bo, CAO Zhu-bing, YU Jie
School of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China
Abstract:
Accurate prediction of exhaust gas temperature (EGT) of aircraft auxiliary power unit (APU) can provide important information for APU health management. Traditional methods have low accuracy in long-term forecasting. A hybrid model of bilateral attention mechanism (BAM) and convolutional neural network (CNN)-gated recurrent unit (GRU) based on characteristic and temporal is proposed, five characteristic parameters with high correlation degree with exhaust temperature were selected to make multivariate prediction of EGT. The introduction of BAM can automatically quantify the correlation between input variables and EGT, and strengthen the expression of historical key information on the predicted output. The introduction of CNN can extract high-dimensional features that reflect the non-stationary dynamic changes of EGT. Experimental results show that the proposed hybrid model achieves good results in single-step and multi-step long time series and multivariate input EGT predictions. Compared with the BAM-GRU model, CNN-GRU model, GRU model, long short-term memory (LSTM) model, support vector machine (SVM) model and back propagation (Back Propagation, BP) model, the prediction accuracy of the hybrid model has been greatly improved.
Key words:  Auxiliary power unit(APU)  Exhaust gas temperature(EGT)  Gated and recurrent (GRU)  Characteristic attention mechanism  Temporal attention mechanism  Convolutional neural network (CNN)