基于多元逐步回归的带钢性能预测模型

Research on prediction model of steel properties based on multiple stepwise regression and data mining

为了挖掘冶金产品关键工艺参数与成品性能间的关系、实现特定牌号性能自动预测,构建了带钢性能预测模型。模型基于多元线性回归的机器学习方法和数据挖掘方法,构建了牌号性能影响因素知识库,通过大量真实生产环境数据,分析其相关性和性能影响因素。实验过程基于2020个预测样本的数据集,其中选择1616个样本作为训练集,404个样本作为测试集,每个样本包括24种属性,对成品性能Y1和Y2进行了单独训练和预测,均方差结果仅为0.0182和2.9371×10^(-5)。实验表明,该性能预测模型预测准确率高,有效命中率达到90%以上,具有良好的应用前景。

In order to explore the relationship between the key process parameters of metallurgical products and the properties of finished products and realize the automatic prediction of the properties of specific grades,a strip performance prediction model was constructed.Based on the machine learning method and data mining method of multiple linear regression,the model constructs the knowledge base of brand performance influencing factors,and analyzes its correlation and performance influencing factors through a large number of real production environment data.The experimental process is based on the data set of 2020 prediction samples,of which 1616 samples are selected as the training set and 404 samples as the test set.Each sample includes 24 attributes.The finished product performance Y1 and Y2 are trained and predicted separately,and the mean square deviation results are only 0.0182 and 2.9371×10^(-5)。Experiments show that the prediction accuracy of the performance prediction model is high,and the effective hit rate is more than 90%,which has a good application prospect.