基于极限学习机的航空发动机主燃油流量预测研究  

Prediction of Aerongine Main Fuel Flow Based on Extreme Learning Machine

作　　者：郭政波 刘振刚 雷杰[1] Guo Zhengbo;Liu Zhengang;Lei Jie(Engine Department,Chinese Flight Test Establishment,Xi’an Shaanxi 710089,China)

机构地区：[1]中国飞行试验研究院发动机所,陕西西安710089

出　　处：《工程与试验》2020年第1期21-23,共3页Engineering and Test

摘　　要：由某型涡扇发动机试飞架次获取训练样本,使用极限学习机(ELM)的方法,通过离线训练建模,得出了发动机主燃油流量逆向预测模型。考虑到输入参数多重线性关系和预测实时性,应用平均影响值(MIV)算法对网络输入参数进行敏感性筛选,最后选取了未参与发动机模型辨识的整个飞行架次的试验数据进行验证。结果表明,在飞行包线内的稳态工况或动态工况下,基于ELM的逆向预测模型的输出与发动机主燃油实际输入基本吻合,具有较高的精度。该模型可用于航空发动机的飞行试验中,对飞行过程中的发动机真实主燃油流量进行监控,以提高安全性。A training sample was obtained from a flight test of a turbofan engine,by using the method of extreme learning machine(ELM)and off-line training modeling,the reverse prediction model of main fuel flow of the engine was obtained.Considering the multiple linear relationship of input parameters and the real-time prediction,the sensitivity of the input parameters of the network was screened by applying the mean influence value(MIV)algorithm.Finally,the test data of the entire flight flight that did not participate in the engine model identification was selected for verification.The results show that under steady-state or dynamic conditions in the flight envelope,the output of the ELM-based reverse prediction model basically coincides with the actual input of the main fuel of the engine,and has high accuracy.The model can be used to monitor the real main fuel flow of an aeroengine in flight test to improve the safety.

关 键 词：神经网络 航空发动机 主燃油流量 极限学习机 平均影响值 

分 类 号：V233.4[航空宇航科学与技术—航空宇航推进理论与工程]