复合正交级数神经网络模型在飞控中的应用

针对现有神经网络逼近快变非线性函数能力低、结构复杂等问题，提出了复合正交多项式模型和一种改进的正匀多项式神经网络模型。这两种模型综合了正交级数和神经网络的结构特点，便于工程实现。理论研究和仿真计算表明，使用此模型构造快变非线性函数时，可使神经网络的节点数成倍减少，神经网络的训练次数成倍降低。飞行器非线性控制的应用表现，此方法可用来解决一般非线性系统的控制问题。

Complex-orthogonal-series-based neural network model and its application

To solve the problems of both low ability to approximate fast varying nonlinear functions and complex structures of present neural networks,a complex orthogonal polynomial neural networks model and an improved complex orthogonal polynomial neural networks model are developed in this paper.The new models are possessed with good approximation to any nonlinearity and simply structured. Theoretical studies and simulating calculations show that when the fast varying nonlinearities are reconstructed by using the new models, both the nodal numbers and the operations of iterative studies of weighted factors of neural networks can be decreased by several times. The new models are used to nonlinear flight control and the results are satisfactory.