一类神经网络观测器的设计及应用

Design and Application of a class of Neural Network Observer

为了减少小波神经网络(Wavelet Neural Network,WNN)的母小波与神经元数,在WNN模型修正的基础上提出了一种能够储存小波上一步信息由自反馈神经元组成的自回归小波神经网络(Self Recurrent Wavelet Neural Network,SRWNN);在分析了这种网络的结构形式后,提出了一类非线性系统的神经网络自适应状态观测器设计方法,并通过引入Lyapunov函数,证明了这种观测器设计方法的正确性;最后,将这种观测器设计方法用于航天器机械手的反演控制,根据SRWNN观测器的估计状态值,应用反演控制理论设计控制器,能够很好地实现系统状态观测,实现无需速度的信号跟踪。

In order to reduce the mother wavelet and the number of neurons of Wavelet Neural Network （ WNN）, a self recurrent wavelet neural network （SRWNN） composed of self feedback neurons which can store the information of wavelet is proposed based on the WNN model. After analyzing the structure of the network, a neural network adaptive state observer design method for a class of nonlinear systems is proposed. By introducing the Lyapunov function, the design method of this observer is proved to be correct. Finally, this observer design method is used in the inversion control of the spacecraft manipulator. According to the estimated state value of the SRWNN observer, the inversion control theory design controller can be used to realize the system state observation well and realize the signal tracking without speed.