智能算法在双辊铸轧过程铸轧力计算中的应用

Casting rolling force calculation for twin-roll strip casting using intelligence method

在双辊铸轧过程中,铸轧力的控制是铸轧过程稳定进行和提高薄带质量的关键.为了控制铸轧力,必须建立铸轧力计算数学模型,本文采用了一种基于贝叶斯方法的前向神经网络训练算法以提高网络的泛化能力,在网络的目标函数中引入了表示网络结构复杂性的惩罚项,融入"奥克姆剪刀"理论,避免了网络训练的过拟合.将上述网络应用于铸轧过程的铸轧力计算,具有很高的计算精度,同时在收敛速度、稳定性和泛化能力方面都优于传统的BP神经网络.

In the process of twin- roll strip casting, rolling force control is the key of the stabilization of the process and the improvement of strips quality. In order to control the rolling force, a mathematic model of rolling force must be established. Bayesian regularization was applied to the training of feedforward neural networks in order to improve their generalization capabilities. Coupled with the ＂Occam＇ s razor＂ theory, a penalty item which could be interpreted as an indication of network complexity, was introduced into the performance function to prevent the occurrence of overfitting. The network was applied in Twin- Roll strip casting rolling force calculation with high precision. Compared with the traditional back - propagation neural network, the Bayesian network has a faster convergence rate, better stability and generalization ability.