摘要
针对当前航空发动机排气温度预测模型精度低的问题,通过分析气路系统特性和飞行数据特点,得出了其状态变化具有渐近连续性的特点,并基于此设计了一种输出层增强的长短期记忆网络模型(OLE-LSTM)。该模型通过在输出层间以及输出层与隐含层间增加信息通路的方法,优化网络对历史输出信息的继承和对当前输入信息的筛选,从而提高模型精度。最后选取执飞不同任务的两个架次发动机排气温度(EGT)数据进行实验,结果表明,相较于标准LSTM、OLE-LSTM模型的平均决定百分比误差由5.232%、7.171%分别减小至1.017%、3.950%,具有更快的收敛速度和更小的稳态误差,有效提高了模型精度。
For the low accuracy of aero-engine exhaust gas temperature prediction model,the characteristics of gas path system and flight data were analyzed,and the conclusion that the state change has asymptotic continuity was obtained.Based on this,an output layer enhanced long short term memory(OLE-LSTM)was designed.The model optimized the inheritance of historical output information by adding information channels between the output layers.At the same time,it optimized the filtering of current input information by adding information channels between the output layer and the hidden layer.Finally,two flights of an aeroengine exhaust gas temperature(EGT)data are selected for experiments.The results show that compared with the standard LSTM,the average percentage error of the OLELSTM model is reduced from 5.232%,7.171%to 1.017%,3.950%.Therefore,OLE-LSTM model has a faster convergence speed and a smaller steady-state error,which effectively improves the accuracy of the model.
作者
张帅
杜军
严智
Zhang Shuai;Du Jun;Yan Zhi(Aeronautics Engineering College,Air Force Engineering University,Xi’an 710038,China)
出处
《电子测量与仪器学报》
CSCD
北大核心
2019年第8期124-132,共9页
Journal of Electronic Measurement and Instrumentation
基金
陕西省自然科学基金(2017JQ6034)资助项目
作者简介
张帅,2017年于湖南大学获得学士学位,现为空军工程大学硕士研究生,主要研究方向为智能数据处理。E-mail:18691854127@163.com.
