基于一种新型动态神经网络的非线性自适应逆控制

Nonlinear Adaptive Inverse Control Using New Dynamic Neural Networks

根据生物神经元的机能,提出了一种具有动态激励函数的新型神经元模型,由此构成的神经网络(DAFNN)应用在非线性自适应逆控制中时只需要确定隐层神经元个数,从而克服了用NARX回归神经网络时需确定输入和输出延时阶数及隐层神经元个数等多个参数的不足。通过对单输入单输出(SISO)及多输入多输出(MIMO)非线性系统的自适应逆控制仿真研究,证实了DAFNN是一种很好的非线性系统建模和控制工具。

When NARX neural networks are used in nonlinear adaptive inverse control, three parameters for archetecture of each network are chosen： the order of input and output delay, the number of hidden nodes. In fact, it is very difficult to choose a set of suitable parameters to guarantee the system stability. A new type artificial neuron, worked as a dynamic mapping, was proposed based on real ones. A new neural network, called DAFNN for its dynamic activation functions in neurons, was built subsequently. There only hidden nodes need to be detemined in DAFNNs. So when DAFNNs are used in nonlinear adaptive inverse control, the stucture of the system become simpler. Then it has a simpler on-line learning algorithm, which improves the convergence of neural networks and stability of the system significantly. Simulation results show DAFNNs are good tools for nonlinear identification and control.