酸轧机组人工神经网络仿真系统  

ANN's Simulation for Sour Rolling Mill

作　　者：吕振辽[1] 华德厚[1] 刘多良[1] 王欣[2] 羌菊兴[2] 宋心棣[2] 

机构地区：[1]东北大学计算中心,辽宁沈阳110004 [2]宝山钢铁集团公司设备部,上海201900

出　　处：《控制工程》2004年第6期571-573,共3页Control Engineering of China

摘　　要：为了提高酸轧机组的控制模型精度,消除模型训练的出错故障,为系统增加了人工干预和在线调整功能。在对原系统进行了深入分析的基础之上,建立了新的酸轧机组人工神经网络仿真系统,并通过网络与生产系统直接相连。制定了数据筛选方法,设计了调整过程算法和离线人工神经网络算法。通过实际应用所产生的数据进行分析,经过离线训练后的全局误差要比初始误差要小得多,误差精度从10-2提高到了10-4。实际应用结果表明了系统的有效性和实用性。To improve the precision of control model in the sour rolling mill and eliminate the faults occurred in the procedure of training model, a method is introduced into the system, which is to adjust the ANN on-line. After thoroughly analyzing the original system, a new simulation of ANN system for the sour rolling mill is erected, and it is directly connected to the production system by TCP/IP. A method of data sort was made and two new algorithms are designed, which are parameter adjustment and off-line ANN training. The global error becomes smaller after off-line training, and the precision is raised to 10^(-4) from 10^(-2) as the practical data showed. The new simulation system proves valid and practical with the results of application.

关 键 词：人工神经网络 轧制力 轧机特性 离线训练 酸轧机组 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]