基于动态RBF神经网络的板形板厚综合控制仿真研究

Simulation research on strip flatness and thickness control based on dynamic RBF neural networks

基于RBF神经网络的特点提出了一种动态调节隐含层隐节点个数的方法,由2部分组成:首先以网络输出数据的均方误差及其变化率为标准来调节隐含层节点的数目,然后调节优化隐含层节点的中心值,根据广义逆矩阵的方法求出输出层权值.所设计的神经网络具有最少的隐含层节点数,提高了学习训练速度,构造了板形板厚综合控制的数学模型,采用新的模型处理方法,用动态RBF神经网络进行控制仿真,取得了理想的结果.

A method to dynamically adjust the number of hidden layer nodes is proposed based on features of the RBFNN, which includes two parts： the first part is to adjust the number of hidden layer nodes based on the mean square error and change rate of network output data, and the second part is to optimize the central value of the hidden layer and find the output layer＇s weights based on the generalized inverse matrix. The newly designed RBFNN has least nodes of hidden layers and higher training speed. A mathematical model for controlling strip flatness and thickness is proposed. Control simulation is executed with dynamic RBF neural network based on new model, receiving an ideal result. ]