Online segmented thickness prediction of hot rolling strip based on IBA-XGBoost

An online segmented thickness prediction algorithm for steel strips based on machine learning is proposed to address issues of strong coupling and low accuracy in existing mathematical thickness models. Firstly, the rolling data are divided into stages of steel biting, accelerated rolling, stable rolling, and steel throwing according to the rolling process. Secondly, an online thickness prediction model with eXtreme gradient boosting (XGBoost) algorithm is established by using segmented data. Then, an improved bat algorithm is applied to optimize the XGBoost model. After that, an adaptive self-learning adjustment method based on the PI closed-loop feedback method is deployed to upgrade the correction speed of the IBA-XGBoost thickness prediction model. Finally, the predicted results are compared with the actual thickness to verify the modeling accuracy. The experimental results indicate that the online segmented thickness prediction model can achieve high accuracy while satisfying time requirements. When IBA-XGBoost uses exit thickness specifications of 3 mm, 4 mm, and 11.45 mm strip rolling data to predict the actual strips thickness, the root mean square error of predicted results is 9.1 μm, 10.3 μm, and 21.8 μm. The results can serve as feedback for the existing automatic gauge control system to further enhance the capability of the thickness control system.