基于LS-SVM的航空发动机气路参数趋势在线预测  

LS-SVM based online trend prediction of gas path parameters of aero engine

作　　者：王旭辉[1] 黄圣国[1] 曹力[1] 施鼎豪[2] 舒平[2] 

机构地区：[1]南京航空航天大学民航学院,南京210016 [2]中国民用航空总局安全技术中心航空安全研究所,北京100028

出　　处：《吉林大学学报（工学版）》2008年第1期239-244,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：“863”国家高技术研究发展计划项目(2006AA12A108);民航基金项目(E0501-MH)

摘　　要：将最小二乘向量机(LS-SVM)应用于航空发动机气路参数趋势在线预测。首先,分析了PATS系统所接收的空地数据链(ACARS)报文的特点,构建用于实时存储和解析的发动机报文并行处理系统,获取在线模型的训练样本;其次,针对实时样本,利用最终误差预报准则(FPE)确定嵌入维数,采用直接搜索法优化LS-SVM建模参数,建立了基于LS-SVM的气路参数预测模型。与航路飞机建立ACARS链接,获取发动机实时气路数据,预测气路参数变化趋势。与已有神经网络预测模型相比,具有更高的一步预测和多步预测精度。该方法解决了发动机预诊断中的数据趋势预测问题。The theory and algorithm of the Least Square Support Vector Machine （LS-SVM） are applied in the trend forecasting of gas path parameters of aero engine. First, the protocol and content of the Aircraft Communication Addressing and Reporting System （ACARS） report are analyzed, the parallel subsystem is constructed for real-time decoding and storing the engine condition report, and the samples for online model training are acquired. Then, the Final Prediction Error （FPE） principle is used to optimize the embedding dimension, the model selection is conducted using Pattern Search Method. An online forecasting model based on the LS-SVM is established. The data link between aircraft and ground is established to acquire the real-time data of the gas path. The online model and the real-time data are used to predict the trend of the gas path parameters. Comparison with the ANN model shows that both one step forecasting accuracy and multi-step forecasting accuracy by the online model are higher than that by ANN model. The proposed online model is suitable for prognosis of gas turbine engine.

关 键 词：航空航天推进系统 航空发动机 在线预测 支持向量机 气路参数 

分 类 号：V231[航空宇航科学与技术—航空宇航推进理论与工程]