航空科学技术2025，Vol.36Issue(10)：73-81,9.DOI:10.19452/j.issn1007-5453.2025.10.009
基于神经网络增强模型的大迎角气动建模与仿真研究
Neural Network-Enhanced Aerodynamic Modeling and Flight Simulation for High-Angle-of-Attack Conditions
摘要
Abstract
When an aircraft operates at high angles of attack,the surrounding flow exhibits significant nonlinear and unsteady characteristics,rendering traditional aerodynamic models inadequate for accurately capturing the aerodynamic behavior under such conditions.To address this limitation,this study introduces an enhanced physics model augmented by neural networks,which effectively integrates the non-linear predictive power of neural networks with the physical accuracy of dynamic derivative models.This hybrid approach significantly enhances the precision and robustness of high-angle-of-attack aerodynamic force predictions.The proposed approach utilizes open-loop harmonic excitation and stall-spin flight data as training inputs for aerodynamic modeling,followed by comprehensive simulation validation.Comparative results demonstrate that the neural network-enhanced physics-based model achieves the smallest prediction errors in aerodynamic coefficients compared to conventional dynamic derivative models and standalone long short-term memory(LSTM)neural networks.Furthermore,the hybrid model exhibits superior accuracy and robustness in spin simulations.This research confirms that the proposed neural network-enhanced physics-based framework substantially improves the fidelity and reliability of aerodynamic predictions for aircraft operating at extreme angles of attack.By bridging data-driven learning and physical principles,this methodology provides a groundbreaking perspective for aerodynamic modeling in aircraft design and development,enabling more accurate characterization of aerodynamic properties in complex flight regimes.Consequently,it establishes a robust foundation for advanced flight dynamics simulations,flight control system design,and flight quality evaluations under critical operational conditions.关键词
大迎角/气动力建模/神经网络/动导数/尾旋仿真Key words
high angle of attack/aerodynamic modeling/neural networks/dynamic derivatives/spin simulation分类
航空航天引用本文复制引用
李镇文,涂良辉,刘吉禹,胡新科,闫超..基于神经网络增强模型的大迎角气动建模与仿真研究[J].航空科学技术,2025,36(10):73-81,9.基金项目
航空科学基金(2023M006056001) Aeronautical Science Foundation of China(2023M006056001) （2023M006056001）
