Strip head thickness prediction during hot rolling process based on clustering feature selection
Thickness is one of the key quality indicators for hot-rolled products.The prediction accuracy of strip head thickness directly affects the control effect of automatic gauge control(AGC),which in turn affects product quality and yield.Due to the complexity and variability of the hot rolling production process,excessive redundancy features seriously affect the effect of thickness prediction modeling.To enhance the accuracy of the prediction model,a method combining hierarchical clustering and mutual information was adopted for feature selection.Strip head thick-ness prediction models were established using deep neural networks(DNN),extreme gradient boosting(XGBoost),support vector regression(SVR),and gradient boosting decision trees(GBDT).Evaluation of the models'generali-zation capability was performed using metrics such as mean absolute error(EMA),mean squared error(EMS),maxi-mum absolute percentage error(EMAP)and coefficient of determination(R2).The results show that the DNN model demonstrates superior precision compared to the others models constructed.The test dataset exhibits EMA,EMS,EMAP and R2 values of 0.015 4,0.000 3,0.004 4,and 0.992 1,respectively,with 97.15％of data having a predic-tion error less than 0.03 mm and a maximum deviation under 0.04 mm.Finally,SHAP method is used for feature analysis,and the influence of related process parameters on strip head thickness is obtained,the high precision pre-diction of strip head thickness is realized.
hot rollingfinishing millclustering feature selectiondeep learninghead thickness prediction
NETL
NSTL
万方数据
