Abstract: Rolling time is a key parameter in the hot rolling production of wide and thick plates.However,due to the complexity and uncertainty of production,it is difficult to accurately preset it in the production preparation stage,which affects the preparation and implementation effect of production operation plan.To solve this problem,based on a large number of wide and heavy plate rolling historical data accumulated in production,the key factors affecting the rolling time and their relationship were analyzed.According to the characteristics of data type and data structure,a two-level decision tree prediction model was proposed to improve the preset accuracy of rolling time.Firstly,the information gain rate of C4.5 was improved based on the dependency between attributes,and the branch nodes were reduced by the level of information entropy.The improved C4.5 classification tree was used to model the nominal attributes in the data.Furthermore,based on Fayyad boundary point decision theorem and support vector machine improved cart algorithm,a regression model for numerical attributes in the classification subset was established.The samples from rolling history data was selected randomly for experiment.The two-level decision tree model was compared with a variety of prediction models to verify the accuracy and robustness of the proposed model.

Key words: rolling time, C4.5 methed, two level decision tree, mixed type data, attribute dependency

摘要： 轧制时间是宽厚板热轧生产的关键参数,但由于生产的复杂性和不确定性,在生产准备阶段很难对其进行精准预设,这会影响生产作业计划的编制以及实施效果。为解决这一问题,着眼于生产中积累的大量宽厚板轧制历史数据,在对影响轧制时间的关键因素及相互关系进行分析梳理的基础上,针对其数据类型和数据结构的特点,提出了两级决策树预测模型,以提高轧制时间的预设精度。首先,基于属性间依赖关系改进C4.5的信息增益率,利用信息熵水平约简分枝节点,将改进的C4.5分类树用于数据中标称属性的建模;进而,基于Fayyad边界点判定定理和支持向量机改进CART算法,对分类子集中数值型属性建立回归模型。从轧制历史数据中随机抽取样本进行实验,将两级决策树模型与多种预测模型对比,验证了所提模型的准确性和鲁棒性。

关键词: 轧制时间, C4.5方法, 两级决策树, 混合类型数据, 属性依赖