神经网络辨识方法及其在轧钢控制中的应用

利用人工神经网络的辨识理论和方法,进行轧制过程数学模型参数的在线辨识与修正.首先对轧制压力模型和温度模型进行分析,得到适于应用神经网络进行辨识和修正的轧制模型函数形式,选择并比较最速下降、递推最小二乘及共轭梯度训练算法,实现了离线的和在线的仿真与应用.仿真结果表明,将人工神经网络应用于轧钢过程的轧制模型辨识可以大大提高模型预报精度.

Model Identification Theory Using Neural Network and Its Application in Plate Rolling Control

A method of identifying and modifying plate rolling model parameters on line with model identification theory using neural network is introduced. Models of rolling force and of temperature were first analyzed to get suitable function styles for identification and modification with neural networks, and several neural network training algorithms, including the one with the steepest gradient, RLS and conjugated gradient algorithm, were chosen and compared. Off line and on line computer emulation and applications were then realized. The results show that the use of neural network in plate rolling process control can greatly improve the precision of model prediction.