一种基于NARX神经网络的纵向飞行动力学建模方法

A longitudinal flight dynamics modeling method based on NARX neural network

针对传统飞行动力学建模工作量大、周期长等问题,提出了一种基于飞行包线离散化和数据驱动的纵向飞行动力学建模方法。以F-16为例,首先分析了纵向运动模态,开展了纵向单脉冲模拟试飞生成特征样本集;然后,利用模糊c-均值算法对样本集进行聚类,选择并设计了综合有效性指标以确定最优聚类数,实现了飞行包线离散化;最后,在子包线内选用训练样本集对NARX神经网络进行训练以建立纵向飞行动力学模型,使用同一子包线内的验证样本集验证模型。结果表明,子包线实现了全覆盖且相邻包线之间边界清晰无重叠,模型输出结果平均相对误差均在3%以内,说明该方法对开展飞机纵向飞行动力学建模是可行的和有效的,为纵向飞行动力学建模提供了一种新思路。

Aiming at the large workload and long period of traditional flight dynamics modeling,a longitudinal flight dynamics modeling method based on flight envelope discretization and data-driven is proposed.Taking the F-16 as an example,the longitudinal motion mode was firstly analyzed,and the longitudinal single-pulse simulation flight test was carried out to generate characteristic sample sets.Then,used the fuzzy c-means algorithm to cluster the sample sets,selected and designed the comprehensive cluster validity to determine the optimal number of clusters,and realized the discreti-zation of the flight envelope.Finally,the training sample set was selected in the sub-envelope to train the NARX neural network so as to establish the longitudinal flight dynamics model,and the validation sample set in the same sub-envelope was used to validate the model.The results show that the sub-envelopes achieve the full coverage and the boundaries between adjacent envelopes are clear and non-overlapping.The average relative error of the model output results is within 3%.It shows that the method is feasible and effective for the modeling of aircraft longitudinal flight dynamics.Also,it provides a new idea for longitudinal flight dynamics modeling.