基于集成特征选择和SVR的热连轧板凸度预测

Crown prediction of hot strip steel based on integrated feature selection and SVR

热连轧板凸度作为评价板形质量的关键指标,具有多变量、非线性、遗传性等复杂特性。传统的热连轧板凸度模型存在机理复杂、理论情况与实际情况存在差异以及模型精度受限等问题。为了解决这些问题,提出了一种基于集成特征选择和支持向量回归的热连轧板凸度预测模型。首先,建立了基于随机森林(Random Forest,RF)、极端梯度提升(eXtreme Gradient Boosting,XGBoost)和梯度提升决策树(Gradient Boosting Decision Tree,GBDT)的集成学习模型,综合应用这些基学习器可以充分挖掘数据中的特征信息;其次,通过集成学习模型对基学习器得到的特征重要性进行加权融合,并根据融合后的特征重要性排序来筛选最具有信息量的模型输入特征,可有效地降低特征维度;然后,采用灰狼优化算法(Grey Wolf Optimization,GWO)来优化支持向量机回归(Support Vector Regression,SVR)预测模型中的参数,其不仅能够消除传统人工参数选择的主观性和盲目性,还能更好地适应数据的特性;最后,将筛选后的特征输入到参数优化的SVR预测模型中,用于对热连轧板凸度进行预测。使用国内某热连轧厂的实际生产数据进行多组试验验证,试验结果表明,模型的预测绝对误差在15μm内的比例超过99%。该预测模型不仅提高了预测精度,还为热连轧板凸度的精确控制和板形质量的改善提供了有力的指导和支持。为解决复杂热连轧生产中的关键问题,以及提高生产过程的可持续性和效率提供了有益的方法和思路。

As a key index to evaluate the shape quality of hot strip steel plate,the crown of hot strip steel plate has the characteristics of multi-variable,nonlinear and hereditary.The traditional crown model of hot continuous rolling plate has some problems,such as complicated mechanism,difference between theory and practice and limited accuracy of the model.In order to solve these problems,aprediction model for crown of hot continuous rolling plate based on integrated feature selection and support vector regression is proposed in this paper.Firstly,an ensemble learning model based on Random Forest(RF),eXtreme Gradient Boosting(XGBoost)and Gradient Boosting Decision Tree(GBDT)is established,and the comprehensive application of these base learners can fully mine the feature information in the data.Secondly,the feature importance obtained by the base learner is weighted by the ensemble learning model,and the most informative input features are selected according to the feature importance ranking after fusion,which can effectively reduce the feature dimension.Then,Grey Wolf Optimization(GWO)is used to optimize the parameters in the Support Vector Regression(SVR)prediction model,which can not only eliminate the subjectivity and blindness of traditional manual parameter selection.It can also better adapt to the characteristics of the data.Finally,the selected features are input into the SVR prediction model with optimized parameters,which is used to predict the crown of hot continuous rolling plate.The experimental results show that the absolute error of the model is more than 99%within 15μm.The prediction model not only improves the prediction accuracy,but also provides powerful guidance and support for the precise control of crown and the improvement of shape quality of hot continuous rolling plate.It provides useful methods and ideas for solving the key problems in complex hot continuous rolling production and improving the sustainability and efficiency of the production process.