全包线内航空发动机参数估计的神经网络实现  

Parameter Estimation for Aero Engine over Full Envelope Using Artifical Neural Networks

作　　者：刘国刚[1] 辛季龄[1] 李璇君 

机构地区：[1]南京航空航天大学动力工程系

出　　处：《南京航空航天大学学报》1999年第3期280-286,共7页Journal of Nanjing University of Aeronautics & Astronautics

摘　　要：介绍一种特殊的前向神经网络——自联想神经网络（Ａｕｔｏａｓｓｏｃｉａｔｉｖｅａｒｔｉｆｉｃｉａｌｎｅｕｒａｌｎｅｔ－ｗｏｒｋｓ，ＡＡＮＮ），然后将发动机参数在全包线、大范围工况下的变化规律与神经网络的非线性映射能力结合起来，开展了将ＡＡＮＮ应用于发动机全包线、大范围工况下参数估计的仿真研究。本文提出的选取测量矢量加入样本集的ＥＭＰ方法，有效地减少了样本集中样本矢量的数目，简化了网络的训练。用ＥＭＰ方法在全包线内仅用７４６组测量矢量作为样本集，在网络训练好后，任选包线内的一工况点作为算例运行发动机模型，所得各参数的稳态估计及动态估计的平均百分比误差＜０．５％。仿真结果表明，上述的参数估值方法是可行的，为进一步实现对发动机控制系统传感器的状态监视和故障诊断打下了基础。First of all,the law of parameter variation of aero engine over full envelope,wide range operating condition is analysed in detail.Choosing autoassociative artificial neural networks (AANN) as an instrument for parameter estimation is briefly introduced.Then combining the law of the parameter variation of the engine over full the envelope,wide range operating condition and the non linear mapping ability of neural networks together, the simulation studies of AANN application to the parameter estimation of the engine over full envelope,wide range operating condition is developed. The EMP method which selects the measured parameter vectors into a sample set can efficiently reduce the scale of sample set and simplify the networks training.Only 746 samples over full envelope are needed. After networks training when arbitrarily selecting an operating point within envelope as an example and running the engine model, the mean percentage error of estimation at steady and transient state is less than 0.5%.The simulation results show that the parameter estimation is feasible.It lays the foundations for further realization of state monitoring and fault diagnosing for sensors of engine control system.

关 键 词：航空发动机 全包线 自联想神经网络 参数估计 

分 类 号：V233.71[航空宇航科学与技术—航空宇航推进理论与工程] TP18[自动化与计算机技术—控制理论与控制工程]