多变量自适应PID型神经网络控制器及其设计方法

Multivariable Adaptive PID-like Neural Network Controller and Its Design Method

提出一种PID型神经网络控制器(PID-like Neural Network Controller,PIDNNC)及其设计方法.基于PID的简单结构和良好性能优势以及神经网络的自调节和自适应的特长,创建一种具有PID结构的多变量自适应的PID型神经网络控制器.该网络控制器的隐含层由带有输出反馈和激活反馈的混合局部连接递归网络组成.通过定义误差函数作为设计目标,采用弹性BP算法,并用变化率以及弹性BP算法中的符号法来处理某些求导关系,获得适于实时在线调整网络权值的修正公式.根据李亚普诺夫稳定性定理推导出确保控制系统稳定的学习速率的取值范围.最后通过实例进一步说明所提出网络控制器的优越性.*

A PID-like Neural Network Controller （PIDNNC） and its design method are proposed. Based on the advantages of simple construction and good property in PID and self-regulation and adaptivity of neural networks, a muhivariable adaptive neural network controller with PID structure is created. It is composed of a hybrid locally connected recurrent network with an activation feedback and an output feedback respectively in the hidden layer. By means of defining error function as the design objective, using resilient back-propagation （BP） algorithm and changing rate and sign in resilient BP algorithm, some differential relations are dealt with. The modified formula for online updating network weights is obtained. Lyapunov stability principle is used to derive the range of learning rate to ensure control system stability. Finally, numerical examples are given to show advantages of the proposed controller.