基于蚁群算法的神经网络冷连轧机轧制力预报  

Neural Network Based on Ant Colony Algorithm for Rolling Force Prediction on Tandem Cold Rolling Mill

作　　者：杨景明[1] 孙晓娜[1] 车海军[1] 刘畅[1] 

机构地区：[1]燕山大学工业计算机控制工程河北省重点实验室,河北秦皇岛066004

出　　处：《钢铁》2009年第3期52-55,共4页Iron and Steel

基　　金："十一五"国家科技支撑计划项目(2007BAF02B12)

摘　　要：为提高冷连轧机轧制力的预报精度和预报速度,用蚁群算法和神经网络相结合的方法进行轧制力预报模型设计。根据轧制原理建立了BP神经网络冷连轧机轧制力预报模型,以网络权值和阈值为自变量,网络预报误差为目标函数,通过蚁群多代运算,找出预报误差全局最小值,再将相应的权值和阈值输入网络进行训练。应用某厂1 450 mm冷连轧机的实测数据进行离线计算的结果表明,该方法能够防止BP网络陷入局部极小点,且收敛速度快,可作为轧制力预报的新方法在实际应用中加以推广。To improve the precision and efficiency of rolling force prediction on tandem cold rolling mill, a neural net- work model combined with ant colony algorithm is presented. The BP （Back Propagation） neural network model for rolling force prediction on tandem cold rolling mill was established according to rolling theory. Taking neural net- work weights and threshold values as decision variables, and neural network prediction error as objective function, the global minimum prediction error could be gotten through multiple generation computation of ant colony. Then training can be done by inputing the corresponding weights and threshold values to the neural network. Using field data on 1450 tandem cold rolling mill, the off-line computation result showed that this method is capable of preven- ting local minimum of BP neural network, and has fast constringency. So it can be generalized in practice as a new method for rolling force prediction.

关 键 词：冷连轧机 轧制力预报 神经网络 蚁群算法 

分 类 号：TG333[金属学及工艺—金属压力加工]