论文部分内容阅读
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。
Combined with the continuous casting production data of a factory, an improved BP algorithm with additional momentum is used to establish a BP artificial neural network prediction model for slab central segregation. The application results show that the prediction accuracy is 90%, which can meet the requirements of segregation prediction accuracy of slab center in continuous casting. The main reason for the bias in the analysis is that there are many hidden layer nodes in the network model, the network structure is complicated, there are many sample learning centers with center segregation level of 1.0 and the noise samples are not completely removed.