基于神经网络的直升机非线性模型辨识研究  

Research on nonlinear helicopter model identification base on neural network

作　　者：王雨润 吴伟[1] WANG Yurun;WU Wei(National Key Laboratory of Helicopter Aeromechanic,NUAA,Nanjing 210016,China)

机构地区：[1]南京航空航天大学直升机动力学全国重点实验室,江苏南京210016

出　　处：《飞行力学》2024年第3期33-39,共7页Flight Dynamics

摘　　要：针对直升机飞行力学建模问题,提出一种基于BP神经网络的非线性辨识方法。在神经网络结构设计中,基于对直升机线性加速度和角加速度产生机理的分析对隐藏层的拓扑形式进行了优化,并结合六自由度欧拉运动方程构建了完整的直升机非线性神经网络模型。在神经网络训练方法方面,建立了基于L-M算法的二步训练方法,对神经网络的开环和闭环结构分两次进行训练。利用UH-60直升机非线性模型生成的扫频激励响应数据集进行神经网络训练与验证。最后,将神经网络模型在各前飞速度下进行配平与线性化,得到了神经网络模型的配平量和气动导数。研究结果表明,所建立的神经网络模型精度高、非线性逼近能力强,并且具有较好的泛化能力。Aiming at the problem of helicopter flight mechanics modeling, a nonlinear identification method based on the BP neural network model was proposed. In the structure design of neural networks, the topology of the hidden layer was optimized based on the analysis of the linear acceleration and angular acceleration mechanisms in helicopters. Furthermore, integrating the six degrees of freedom Euler motion equations, a complete non-linear neural network model for helicopters was constructed. In the aspect of neural network training method, a two-step training method based on L-M algorithm was established, and the open loop neural network and closed loop neural network were trained respectively. The neural network was trained and validated using a dataset of sweep excitation responses generated by the UH-60 helicopter's nonlinear model. Finally, the neural network model was trimmed and linearized at various forward flight speeds, and the trim quantity and aerodynamic derivative of the neural network model were obtained. The research results show that the established neural network model has high accuracy, strong non-linear approximation capabilities, and good generalization abilities.

关 键 词：直升机 BP神经网络 L-M算法 

分 类 号：V275.1[航空宇航科学与技术—飞行器设计]