taptap点点手机网页

• 民用航空 • taptap点点手机网页版    下一篇

基于特征选择和极限学习机的发动机性能预测

徐建新,岳敏骐   

  1. (taptap下载安装安卓航空工程学院,天津300300)
  • 收稿日期:2014-12-29 修回日期:2015-03-16 出版日期:2016-02-20 发布日期:2016-04-13
  • 作者简介:徐建新(1967—),男,江苏苏州人,教授,博士,研究方向为飞机结构疲劳强度和复合材料力学.

Aeroengine performance prognosis based on feature selection and extreme learning machine

XU Jianxin,YUE Minqi   

  1. (College of Aeronautical Engineering,CAUC,Tianjin 300300,China)
  • Received:2014-12-29 Revised:2015-03-16 Online:2016-02-20 Published:2016-04-13

摘要:

利用PW4000 发动机的实时监控数据建立数据库,以平均影响值为评价标准进行了特征选择,筛选出8 个特征参数作为模型输入,训练基于极限学习机算法的单隐层神经网络,建立了排气温度预测模型。用PW4000发动机的运行数据进行了模型验证,与误差逆传播算法进行了对比,并用发动机水洗恢复之后的数据进行了拓展性研究。测试结果显示利用平均影响值进行特征选择结果可信度较高,极限学习机的运算速度快于误差逆传播算法,有利于多次运算充分发挥其优势袁整个算法误差较小,修正后的模型具有良好的拓展性。

关键词: 航空发动机, 实时监控数据, 平均影响值, 特征选择, 极限学习机, 神经网络

Abstract:

For predicting exhaust gas temperature of aeroengine,a modeling method is performed. Original database is derived from real-time monitoring data of engine monitoring system of PW4000 engine. The mean impact value (MIV)is as taken evaluation criteria for feature selection,screening 8 features as input variables. Extreme learning machine (ELM)algorithm for single-hidden layer feedforward neural network is applied for modeling,After the neural network is tested by monitoring data of the same engine,and model's extended performance is tested by data of that same engine after water wash. Result shows that training time of ELMis much less than that of back propagation algorithm,which is efficient for repeated training for reducing error,and the whole modeling method has satisfactory scalability.

Key words: aeroengine, real-time monitoring data, mean impact value, feature selection, extreme learning machine, neural network

中图分类号: 

Baidu
map