一种PID型模糊神经网络控制器

A Kind of PID-Typed Fuzzy Neural Network Controller

为了使一种基于两维控制规则基的PID型模糊控制器具有参数在线学习功能,提出了一种包含一个自回归神经元的五层模糊神经网络,并根据梯度下降法,给出了它各层权值的修正算法,该网络可以在反馈控制系统中作为一个自学习控制器来使用。最后,根据有关定理,给出并证明了该网络各层权值学习速率的收敛准则。

In order to realize the self-learning ability of a kind of PID-typed fuzzy controller designed only using two-dimension rule base, we design a five-layer fuzzy neural network which contains a self-recurrent neuron and propose the weights correcting method of all its layers on the gradient descent algorithm. This network can be used as a self-learning controller in the feedback control system. It is also given and proved the convergent theorem of the weights learning rate in the network based on the relative theorem.