基于改进BP算法的连铸板坯中心偏析预报模型

Central Segregation Prediction Model of Continuous Casting Slab Based on Developed BP Neural Network Algorithm

结合某厂连铸生产数据,采用带有附加动量项的改进BP算法,建立了连铸板坯中心偏析的BP人工神经网络预测模型。应用结果表明,其预测准确率为90%,可满足连铸生产中对铸坯中心偏析预报精度的要求。分析导致预报偏差的主要原因是,网络模型隐含层节点较多、网络结构复杂、中心偏析等级为1.0的样本学习次数较多和噪音样本剔除不彻底等。

Central segregation and internal crack decrease the properties of slab and rolling plate badly. In this paper, on the basis of production records for some steel corporation, an improved BP algorithm with the additional momentum term is put forward, which is used to develop a BP neural network model to predict the central segregation of continuous casting slab. The results showed that, prediction accuracy of central segregation model is 90%, which meet the demand of prediction of central segregation well. The main reasons for prediction errors include excess hidden layer numbers in the neural network structure, excess sample numbers of central segregation defect at 1.0 level and noise samples being not eliminated absolutely.