Study and application of data-driven rolling force prediction model for hot strip mills
The current hot strip rolling gradually presents the characteristics of variety diversity and process com-plexity.Due to the consideration of fewer factors,the traditional rolling force prediction models gradually reveal some defects and deficiencies,and cannot meet the requirements of high-precision and high-performance product con-trol accuracy.Therefore,for a 2 250 mm hot rolling finishing mill in China,an intelligent rolling force model based on mechanism model and gradient boosting decision tree algorithm was developed using data mining technology and intelligent algorithms,combined with mechanism models.The model had comprehensively considered various fac-tors,enabling the acquisition of plan sheet data in advance and targeted training for the steel grades to be rolled,thereby enhancing the prediction accuracy for small samples.Additionally,self-training and closed-loop control technologies had been developed,allowing for deployment and application in various environments and achieving au-tomatic closed-loop control.After applying the model on-line,the long genetic prediction accuracy error of rolling force was controlled within 5％,and the single calculation time took less than 10 ms.The results show that the model has fast response speed,high calculation accuracy and good calculation stability,which can meet the require-ments of rolling force accuracy control under changing steel grades and working conditions,thereby improving the stability of strip rolling and the control accuracy of head thickness,and enhancing product competitiveness.
hot rolled steel striprolling force predictionmechanism modelGBDT algorithmmodel application
NETL
NSTL
万方数据
