基于长短时记忆神经网络的非定常气动力建模方法  

Unsteady aerodynamics modeling method based on long short-term memory neural network

作　　者：何磊[1] 张显才 钱炜祺[1] 张天姣[1] HE Lei;ZHANG Xiancai;QIAN Weiqi;ZHANG Tianjiao(China Aerodynamics Research and Development Center,Mianyang 621000,China)

机构地区：[1]中国空气动力研究与发展中心,四川绵阳621000

出　　处：《飞行力学》2021年第5期8-12,共5页Flight Dynamics

摘　　要：针对飞机大迎角过失速机动过程中的非定常气动力高精度建模需求,提出了一种基于长短时记忆(LSTM)神经网络的非定常气动力建模方法。以三角翼大迎角非定常气动特性为研究对象,建立了基于LSTM神经网络的非定常气动力模型,实现了对升力系数、阻力系数和俯仰力矩系数的预测。研究结果表明,基于LSTM的非定常气动力模型收敛速度快,模型预测结果与真实试验结果符合性较好,模型预测精度优于基于循环神经网络的非定常气动力模型,并且具有良好的泛化性能。To meet the high-precision modeling requirements of nonlinear and unsteady aerodynamic force during aircraft post-stall maneuver at high angle of attack, an unsteady aerodynamics modeling method based on long short-term memory(LSTM) neural network was proposed. The LSTM model of unsteady aerodynamic characteristic of delta wing at high angle of attack was built to predict the aerodynamic coefficients of lift, drag and pitching moment. The results show that the unsteady aerodynamics model based on LSTM converges quickly in training process, and the predicted results of the model fit well with the true experimental results. The accuracy of predicted results of unsteady aerodynamic model based on LSTM is better than that based on recurrent neural network, and the model has a good generation performance.

关 键 词：长短时记忆 机器学习 非定常 气动力建模 大迎角 

分 类 号：V211.4[航空宇航科学与技术—航空宇航推进理论与工程] TP183[自动化与计算机技术—控制理论与控制工程]