RBF网络优化设计及在轧机轧制力预报中的应用  

Optimized Design of RBF Network and Its Application in Rolling Force Prediction

作　　者：董敏[1] 刘才[1] 李灵锋[2] 

机构地区：[1]燕山大学轧机研究所,河北秦皇岛066004 [2]河北建材职业技术学院信息机电系,河北秦皇岛066004

出　　处：《钢铁》2005年第11期34-36,61,共4页Iron and Steel

基　　金：国家自然科学基金资助项目(50375135);河北省自然科学基金资助项目(E2005000323)

摘　　要：将Hough变换用于RBF神经网络的参数确定中,以自适应地确定RBF网络的隐层节点数和径向基函数的中心值,使得网络结构及参数得到优化,提高了RBF神经网络的收敛速度和泛化能力。将此改进的RBF神经网络应用于冷轧过程轧制力的预测中,试验结果表明经过优化设计的网络模型具有高的输出精度,使得轧制力预报这种受多种因素影响的复杂系统的输出预报问题得到了很好的解决。Hough transform was applied to select parameters of RBF networks, so as to decide the number of hidden layers and the center value of Gauss function adaptively. Networks convergence speed and generalization ability were enhanced by parameter optimization. The improved networks were applied in prediction of rolling force on cold strip mill. Experiment results indicated that the optimized network has short computation time, high accuracy and is suitable for rolling force prediction, which is influenced by many factors.

关 键 词：HOUGH变换 RBF神经网络 轧制力 预报 

分 类 号：TG333[金属学及工艺—金属压力加工]