一种自构建神经网络的内模控制方法

An Internal Model Control Method Based on Self-Constructing Neural Network

针对非线性过程,提出了一种基于自构建神经网络的内模控制方法(Internal Model Control,IMC)。采用自构建算法实现神经网络的结构学习和参数学习,在被控过程内部模型和控制器模型的辨识过程中,该网络能够根据给定的判定条件自动增加神经元节点,以满足辨识精度的要求;为了防止网络学习过拟合,基于灵敏度方法对神经网络隐层节点进行修剪删除;网络的参数学习采用梯度下降法。自构建算法可以有效地避免普通神经网络内模控制方案中网络结构难以确定的问题,仿真结果表明,该控制系统有良好的跟踪性、鲁棒性和抗干扰性。

A novel algorithm on internal model control （IMC） based on self-constructing neural network （NN） is proposed for the non- linear process in this paper. The structure learning and parameters learning of the neural network were realized by self-constructing algo- rithm. In the identification process of the internal model and the controller, the neural network can automatically increase the nodes to meet the identification accuracy requirements. Moreover, in order to prevent the over-fitting of neural network learning, the hidden layer nodes can be pruned based on the sensitivity method. In addition, parameters learning adopt the gradient descending method. Compared with conventional NN-IMC method, this scheme can effectively avoid the problem of network structure is difficult to determine. The sim- ulation result shows that the control system has a good target tracking performance, disturbance rejection properties and robustness sim- ultaneously.