极限学习机在中厚板轧制力预报中的应用

摘要：轧制力是影响中厚板厚度精度和板型的关键因素。兴澄特钢中厚板轧机二级模型采用传统Sims公式计算轧制力，精度较低。为提高轧制力预报精度，首先基于大量历史生产数据，通过主成分分析法对影响轧制力的因素进行处理和分析，选出权重较大的影响因子；其次选取现场代表钢种进行热模拟压缩实验，在此基础上提出基于极限学习机(ELM)的综合神经网络轧制力预报模型，即先通过化学成分计算出基准变形抗力，再将其作为轧制力神经网络输入变量进行轧制力预报。建模采用10折10次交叉验证确定最佳网络隐层节点数，并用现场实际生产过程数据对网络进行训练与测试。综合神经网络模型投入现场生产，轧制力预报相对误差±10%以内占比提高15.61%，钢板头部厚度命中率提高1.9%。

Application of ELM to predict plate rolling force

Rolling force is a key factor affecting the thickness accuracy and shape of the plate. The traditional Sims formula was used to calculate the rolling force in the L2 model of Xingcheng plate mill with low precision. In order to improve the accuracy of rolling force prediction, firstly based on a large number of historical production data, the principal component analysis method was used to process and analyze the factors affecting the rolling force, and the influential factors with larger weight were selected. Secondly, the representative steel grades were selected for thermal simulation compression experiment, on this basis, the comprehensive neural network rolling force prediction model based on the extreme learning machine (ELM) was proposed, which firstly calculates the basic yield stress by chemical composition, and then uses it as the input variable of rolling force neural network to carry out rolling force prediction. 10 fold 10 times cross validation was used to determine the optimal number of hidden layer nodes, and the actual production process data was used to train and test the network. The integrated neural network model was put into real production, the relative error of rolling force prediction within ±10% was increased by 15.61%, and the hit rate of plate head thickness was increased by 1.9%.