基于神经网络的连铸坯表面纵裂预测  

CC-slab surface longitudinal crack prediction by using neural networks

作　　者：王悦新[1] 邱以清[1] 刘相华[1] 

机构地区：[1]东北大学轧制及连轧自动化国家重点实验室,沈阳110004

出　　处：《材料与冶金学报》2009年第1期21-24,32,共5页Journal of Materials and Metallurgy

基　　金：国家自然科学基金资助项目(50574018)

摘　　要：针对铸坯质量预报问题,利用人工神经网络中的BP算法建立原始化学成分和连铸生产的主要工艺参数与产品最终质量之间的关系,并开发出专门的应用软件.软件共分3部分:数据处理部分、人工神经网络训练部分、运用成熟网络预报部分.该预测方法的特点是直观、方便、稳定.数据均从稳定生产的现场取得.采用神经网络对D32-1钢的铸坯质量进行预报,经过上千次训练后,产品质量的预报值与实际值拟合良好,预报结果的相对误差很小.Dealing with casting slab quality prediction, the BP algorithm of ANN was used to build the relation between original chemical composition together with the main craftwork parameters of continuous casting and terminal quality of the products. This software has three parts： data processing, data training part of the ANN, predicting part by using the mature ANN. The characteristic of this software is visible, convenient and stable. Data are acquired from the stable production line. There is an example of predicting CC - slab quality of D32 - 1 steel using this software. On the basis of thousand times training, the predicted results are in a good agreement with the measured values, relative error in the predicted results is small.

关 键 词：人工神经网络 BP算法 纵裂 预测 

分 类 号：TF777[冶金工程—钢铁冶金] TP183[自动化与计算机技术—控制理论与控制工程]