神经网络在发动机自适应建模中的应用研究

Application of Neural Networks to Areoengine Adaptive Modeling

提出了一种新的基于神经网络的发动机自适应实时模型的建模方法。建模的思想是认为发动机的任何非额定工作都将导致其输出参数的变化,因而可以把这些参数偏离正常工作参数值的变化量,也就是输出偏离量,用来表征发动机的非额定工作情况。把它们作为增广的状态变量,设计卡尔曼滤波器对其进行最优估计,然后用这些输出偏离量的估计值,通过由BP神经网络训练出来的可测输出偏离量与未测输出偏离量的映射关系来校正机载发动机模型的计算输出,使之与真实发动机的输出一致,从而使实时机载模型获得对任何发动机非额定工况的自适应能力。

In this paper a new method based on neural networks is proposed for aeroengine adaptive real time modeling.It is thought that output data of aeroengine will bias their nominal values in any cases of non-nominal work.So the biases of output data can be used to represent the non-nominal work conditions of the aeroengine and to constitute the augment state variables.Kalman filter is designed to estimate these biases.Then the unmeasured outputs of the onboard model of the areoengine can be modified by the estimation of the unmeasured biases which are obtained through the mapping relation constructed between the measured output biases and unmeasured ones and trained by the BP neural networks offline.After modification,outputs of onboard model are the same as those of the real aeroengine,and real time onboard model has the abilities of adaptation.