神经网络模型在TDNS准则中的应用研究

时域Neal-Smith（TDNS）准则源于频域Neal-Smith（FDNS）准则和阶跃目标跟踪准则，被广泛应用于纵向驾驶员诱发振荡（PIO）趋势的预测研究。在该准则采用的人机闭环系统中，广泛运用结构简单且能反映飞行员基本行为特征的拟线性驾驶员模型。但在实际情况中，驾驶员的行为通常具有高度的非线性，为了更精确地描述驾驶员的非线性特性，采用神经网络方法建立驾驶员模型。利用某机的试飞数据对神经网络进行了训练和测试，生成了神经网络模型。针对TDNS预测准则，建立了相应的仿真模型，通过仿真运算，得到了该飞机的P10预测结果。研究表明：相对于传统的驾驶员模型，神经网络模型能够更好与真实驾驶员相匹配；在TDNS评估准则中运用神经网络模型具有实际的意义。

Research on Neural Network Model Used in TDNS Criterion

Time domain Neal-Smith (TDNS) criterion is widely used in predicting longitudinal pilot induced oscillation (PIO) which is originated from frequency domain Neal-Smith (FDNS) criterion and step target tracking criterion. In the closed loop system of this criterion, quasi-linear pilot model is extensively used because of its capability of reflecting basic characteristics of pilot’s behavior and its simple structure. But in reality, the pilot’s behavior is usually highly nonlinear. In order to describe the nonlinear characteristics more accurately the neural network method is used to establish the pilot model. Taking an aircraft as an example, flight test data is used to train and test the neural network, and then a model of neural network is created. Based on TDNS criterion, the simulation model is established. Through computer simulation, the prediction results of the PIO trends are obtained. Research shows that compared with the traditional pilot model, neural network model can match with the real driver better. Therefore, putting neural network model into TDNS assessment criterion is of actual significance.