基于状态感知的航空发动机变基线模型建模方法研究

Modeling Method of Aeroengine Variable Baseline Model Based on State Perception

关键词:  航空发动机  机载模型  飞行数据  状态感知  高斯混合模型  变基线模型

Key words:  Aero engine  On-board model  Flight data  State perception  Gaussian mixture model  Variable baseline model

摘要:

机载模型是先进航空发动机控制方法的基础，基线模型作为机载模型的重要组成部分，其建模准确度决定了机载模型的精度。针对传统单一基线模型在局部飞行包线精度高，而难以用于发动机全包线、全状态稳态性能预测的问题，提出了一种基于状态感知的发动机变基线模型建模方法。首先在小波变换滤波的基础上，提出基于状态感知的最优稳态数据筛选阈值计算方法，以减少稳态数据的错选或遗漏；其次，提出基于高斯混合模型（GMM）的变基线模型建模方法，利用GMM实现飞行数据自主聚类，并结合回归分析法，构建全包线、全状态的高精度变基线模型。仿真结果表明：本文提出的稳态数据筛选方法能有效避免数据错选或遗漏，相比于常规的单一基线模型，所提出的变基线模型可使高、低压转子转速的相对均值误差分别减小45%，30%以上。该方法能显著提升基线模型精度，同时实现了稳态数据自动化提取，避免了过多依赖人工经验且难以获得最优阈值的问题。

Abstract:

On-board model is the basis of advanced aeroengine control method. Baseline model is an important part of airborne model， and its modeling accuracy determines the accuracy of on-board model. Aiming at the problem that the single baseline model has high accuracy in the local flight envelope and is difficult to be used in the prediction of engine full envelope and full state steady-state performance， a variable baseline model modeling method of aeroengine based on state perception is proposed. Firstly， based on wavelet transform filtering， a calculation method of optimal steady-state data screening threshold based on state perception is proposed to reduce the wrong selection or omission of steady-state data. Secondly， a variable baseline model modeling method based on Gaussian mixture model （GMM） is proposed. GMM is used to realize the autonomous aggregation of flight data， and combined with regression analysis， a high-precision variable baseline model with full envelope and full state is constructed. The simulation results show that the proposed steady-state data screening method can effectively avoid data misselection or omission. Compared with the conventional single baseline model， the proposed variable baseline model can reduce the relative mean error of high and low pressure rotor speeds by more than 45% and 30%， respectively. The proposed method significantly improves the accuracy of baseline model， realizes the automatic extraction of steady-state data， and avoids the problem of relying too much on manual experience and difficulty of obtaining the optimal threshold.