基于连铸生产大数据的热轧卷质量预测模型

板坯热送热装或连铸连轧技术逐渐被越来越多的钢厂采用，然而其进一步发展受到连铸坯质量的制约。故针对连铸坯缺陷的有效判断可避免存在质量问题的铸坯进入轧制环节，从而降低额外的能耗。基于对铸坯质量的在线检测困难这一问题，从生产大数据的角度建立了板坯热轧卷的质量预测模型。首先根据正常与缺陷产品高度不平衡的数据特点，提出了相关性分析、不平衡数据随机分类与主成分数据降维三者相结合的数据预处理方法，随后选择GA-BP神经网络算法构建了针对低碳钢、包晶钢和中碳钢的热轧卷质量预测模型。预测模型具有较高的准确率，其中低碳钢模型总体预测准确率达到94.7%,缺陷预测准确率为82.8%;包晶钢模型总体预测准确率达到93.3%,缺陷预测准确率为87.5%;中碳钢模型总体预测准确率为85.4%,缺陷预测准确率为87.3%。最后，基于Python语言编写了热轧卷质量在线预测软件，可对热轧卷质量进行实时预测，方便快速地溯源缺陷发生原因。

Quality prediction model of hot rolled coil based on big data of continuous casting process

Hot charging or continuous casting and rolling technology of slabs have been gradually adopted by more and more steel enterprises, but its further development is restricted by the quality of slab. Therefore, the effective judgment of continuous casting slab defects can prevent the slab with quality problems from entering the rolling process, so as to reduce the extra energy consumption. Based on the difficulty of online quality detection for slabs, the quality prediction models of hot rolling coil of slabs were established from the perspective of big data production. Firstly, according to the highly unbalanced data of normal and defective products, the data preprocessing method combining correlation analysis, random classification of unbalanced data and dimensionality reduction with principal component analysis was proposed. Then, the GA-BP neural network algorithm was selected to construct the hot rolled coil quality prediction models for low carbon steel, peritectic steel and medium carbon steel, respectively. The results showed that the prediction model has a high accuracy, and the overall prediction accuracy of the low carbon steel model reaches 94.7%, and the defect prediction accuracy is 82.8%. The overall prediction accuracy of the peritectic steel model is 93.3%, and the defect prediction accuracy is 87.5%. The overall prediction accuracy of medium carbon steel model was 85.4%, and the defect prediction accuracy was 87.3%. Furthermore, an online prediction software for hot rolled coil quality was designed based on Python language, which could predict the quality of hot rolled coil in real time and trace the causes of defects conveniently and quickly.