大飞行包线控制律的神经网络调参设计

对于具有大飞行包线的现代电传飞行控制系统,给出了一种3层BP(Back Propagation)神经网络代替传统的根据动压、高度或马赫数单一调参的控制律,解决了大飞行包线内控制参数复杂、单一调参无规律性、在全飞行包线上特别是在平衡点之间的飞行状态的稳定性无法保证等问题.通过离线训练,得出了一组隐层只有6个节点的网络结构参数,输出为各平衡点设计的最优鲁棒反馈增益.利用该网络实现了大飞行包线的根据高度和马赫数的双参数增益调参.仿真表明,利用该神经网络可以保证在所有平衡点上原设计的最优反馈增益不变,响应过程不变,同时可以细化平衡点之间的控制参数,在较大建模误差(约50％)和平衡点间也可以具有较好的控制效果.

Neural network gain scheduling design for large envelope curve flight control law

A 3-layer BP(back propagation) network is developed for the gain scheduling flight control law in a large flight envelope curve for a flight by wire flight control system. It can replace the traditional gain scheduling according to the altitude, Mach number or dynamical pressure which is only one-parameter adjusting and then difficult to find the adjust rule and so the stability in hole flight envelope curve, specially for those flight conditions between the designed points, can't be guaranteed. The design procedure included choosing a BP network with middle layer of 6 points only and its parameter train. The outputs are the designed optimal robust 8 feedback gains. This network realized double parameters (altitude and Mach) gain adjusting. The illustration shows the neural network can give the same control parameters at the designed flight conditions and then the response can be the same good as before. It gets a good control effect at other un-designed flight conditions, also for the system with about 50% dynamical modeling error. It gives more finely carve up result of the control parameters between all flight conditions and shows a strong extrapolate ability.