Interval prediction and control of front-end bending in hot rolling

The point prediction of slab front-end bending can provide a specific quantitative value of bending as the prediction result, but cannot accurately give the probability and fluctuation range of front-end bending in hot rolling. In this paper, an innovative full-process collaborative platform that incorporates an ensemble interval prediction method paired with a control strategy is proposed. This approach effectively represents the uncertainty associated with front-end bending prediction points in the form of probability distributions and establishes corresponding evaluation criteria with proven reliability. The machine vision technology is utilised to collect rolling data in real-time, which is subsequently processed to provide input data for the prediction model. A non-dominated sorting genetic algorithm is adopted to optimise the extreme gradient boosting model for the front-end bending prediction. Meanwhile, the prediction interval of the front-end bending is obtained by combining a non-parametric statistical analysis. Finally, based on the prediction interval combined with the proposed control strategy and evaluation criteria, an real-time regulation system of the hot rolling slab front-end bending is achieved. The online application results show that the method proposed in this paper can make timely and accurate responses to the fluctuations of slab front-end bending, effectively improving the control effect and product quality.